UWE has obtained warranties from all depositors as to their title in the material deposited and as to their right to deposit such material.

Transcription

1 Russell, J. J. (2016) Development of generic methods for the analysis and purification of polar compounds by high performance liquid chromatography. PhD, University of the West of England. Available from: We recommend you cite the published version. The publisher s URL is: Refereed: No (no note) Disclaimer UWE has obtained warranties from all depositors as to their title in the material deposited and as to their right to deposit such material. UWE makes no representation or warranties of commercial utility, title, or fitness for a particular purpose or any other warranty, express or implied in respect of any material deposited. UWE makes no representation that the use of the materials will not infringe any patent, copyright, trademark or other property or proprietary rights. UWE accepts no liability for any infringement of intellectual property rights in any material deposited but will remove such material from public view pending investigation in the event of an allegation of any such infringement. PLEASE SCROLL DOWN FOR TEXT.

2 Development of generic methods for the analysis and purification of polar compounds by high performance liquid chromatography Joseph Jonathan Russell A thesis submitted in partial fulfilment of the requirements of the University of the West of England, Bristol for the degree of Doctor of Philosophy This research programme was carried out in collaboration with GlaxoSmithKline, Stevenage, UK Faculty of Health and Applied Sciences, University of the West of England 16 May 2016

3 Abstract Generic methods were developed using different columns for analysis and purification of hydrophilic compounds by hydrophilic interaction chromatography (HILIC). Mobile phases were investigated in detail, and across each column chemistry tested (BEH Amide, Atlantis bare silica, ZIC-HILIC and Cogent Hydride), salt-buffered mobile phase offered good to excellent peak shape for acids, bases and neutral solutes with a range of hydrophilicities. Additionally, cation exchange occurred on the bare silica column even when rubidium nitrate was added to the mobile phase, which should block all cation exchange sites. Measurement of mobile phase ph in hydroorganic solvent (ACN-water mixture with buffer) better represented the environment solutes experience on column than fully-aqueous ph measurement. The performance of HILIC with Charged Aerosol Detection (CAD) was evaluated with a hydrophilic acid, a hydrophobic base and a hydrophilic neutral solute; limits of detection and quantitation were 1-3 ng and 5-9 ng on column, respectively. This compared favourably to literature values for other universal detectors. HILIC-CAD was further investigated by flow injection analysis (FIA) using 29 solutes containing acids, bases and neutrals. HILIC and CAD had excellent compatibility: peak areas were double compared to reversed-phase conditions, response was reasonably uniform for 21 non-volatile solutes considering the Page 2 of 246

4 solutes diversity. HILIC-CAD was viable for retention and detection of highly hydrophilic species without chromophores: salts, sugars and amino acids. Salts travelled down the column as independent cations and anions. Resolution of sugars and amino acids was challenging and was incomplete due to project time constraints. Generic methods were developed on an analytical system in the labs of the industrial collaborator and applied to purifications on wide-bore columns at scaled-up flow rates (21mm id, 20mL/min prep vs. 4.6mm id 1 ml / min analytical analytical). A standard prep system was capable of usable productivity using HILIC with 1mL injections (22 mg of crude purified per hour) and use of At-Column Dilution enhanced this around 10-fold with scope for 4mL injections (223 mg of crude purified per hour). Page 3 of 246

5 Acknowledgements I thank Professor David McCalley for his support throughout the project. His expertise in HPLC and guidance in preparing papers worthy of high-impact journals has been invaluable. I would also like to thank Dr James Heaton, who has given me reasoned critique with loyal friendship. I offer thanks to Dr Bob Boughtflower and Tim Underwood at GlaxoSmithKline for preparing the bid for my CASE studentship in collaboration with Professor McCalley. The intense HPLC training they gave me at the start of my project and supervision of my prep work in a second placement set me in good stead, with their balance of sound theory and hard-nosed pragmatism. I thank Simon Readshaw at GlaxoSmithKline for supporting my project from start to finish. I thank ThermoFisher Scientific for the loan of UHPLC instrumentation and Charged Aerosol Detectors, especially Dr Frank Steiner, Dr Tony Edge and Dr Norman Ramsey. Thanks also to the Chromatographic Society for providing financial support to attend meetings in the UK and International HPLC conferences in New Orleans, USA and Geneva, Switzerland, in particular Dr John Lough. I hope to continue as a UK researcher who produces chromatography publications and play a future role in this field. I would like to thank the Royal Society of Chemistry for providing financial support to attend meetings and the opportunity to learn management skills by becoming an Early Career Network Representative in the Bristol & District local section. I thank Colin Chapman for bringing me onto the local committee and Giovanni Depietra and Niamh Brannelly for collaborating in the revival of the Early Career Network in our local area. Finally, I thank Professor John Hart for being my second supervisor. As part of such a large and eminent supervisory team, John has been a friend throughout the project. This project was funded by an EPSRC CASE studentship award with GlaxoSmithKline (GSK). Initial HPLC training was done at GSK Stevenage during a three week placement Oct-Nov 2012 and a purification study in a placement Sep-Oct A UHPLC system with CAD detection was provided by ThermoFisher Scientific, with beta-testing of a new CAD Veo detector (Chapter 5). Page 4 of 246

6 Dedication This thesis is dedicated to my wife Lisa. She keeps me looking up; with her I can see that the chasm is but a crack. Page 5 of 246

7 Contents Acknowledgements... 4 Dedication... 5 Contents... 6 Chapter High-Performance Liquid Chromatography HPLC separation Polarity and Hydrophilicity HPLC of polar pharmaceuticals Ion pair chromatography Hydrophilic Interaction Chromatography HPLC detection Charged Aerosol Detection Purification of polar pharmaceuticals Focused Gradient Liquid Chromatography and At-Column Dilution Objectives Chapter Instrumentation a. HILIC buffer experiments b. Charged Aerosol Detector experiments c. HILIC Generic and focused method development d. HILIC prep experiments Conditions a. Injection Chemicals and reagents Probe solutes Chapter Abstract Introduction Experimental Results and discussion Buffer and solute properties Page 6 of 246

8 3.2 Initial studies to establish a generic HILIC mobile phase buffer: performance of four different phases with three mobile phase buffers Detailed studies to elucidate phenomena responsible for results in Comparison of performance of four different stationary phases with three different buffers Effect of mobile phase water concentration and buffer on retention and peak shape Causes of poor peak shape for cationic solutes in formic acid Effect of buffer salt concentration and salt cation on retentionof cationic compounds Conclusions Chapter Abstract Introduction Experimental Chemicals and reagents Equipment and methodology Results and discussion Detection limits (HPLC) Calibration curves (HPLC) Response universality and uniformity Flow injection analysis Effect of solute salt composition on response (HPLC; FIA) Response Uniformity (FIA)-dependence on solute and mobile phase buffer Effect of solute volatility on response Effect of organic modifier (FIA) Effect of elevated temperature (FIA) Analysis of salts (HPLC) Conclusions Chapter Abstract Introduction Experimental Chemicals and reagents Equipment and methodology Results and discussion Page 7 of 246

9 3.1 Viability of sugar analysis using an alternative stationary phase (BEH Amide) and CAD Veo detection Comparison of Ultra and Veo CAD Noise in buffered and unbuffered mobile phase Analysis of Simple Sugars in Beer and Cider Analysis of underivitised Amino Acids Conclusions Chapter Abstract Introduction Theory Loadability Peak shape Solubility Productivity Experimental Chemicals and reagents Apparatus and methodology Results and discussion Choosing separation conditions with alternate selectivity Changing the selectivity using the organic solvent Changing the selectivity by changing the stationary phase in generic HILIC methods Focused analytical zone methods Changing the selectivity using RPLC toolbox methods at low ph and high ph Sample preparation using different injection solvents to improve analyte solubility Purification by HILIC Generic vs. focused methods on an Atlantis column Generic vs. focused methods on a BEH Amide column Purification by HILIC using an At-Column Dilution (ACD) system Preparative productivity Purification of zwitterion(s) Possible explanations for poor ACD performance using DMSO or TFA Characterising of the effect of DMSO on peak shape using an analytical system Loss of resolution in ACD with 10% TFA diluent Page 8 of 246

10 5. Conclusion Chapter Overall Conclusion Further Work References Appendix I Figures, Tables, Equations, Symbols and Abbreviations I.1 List of Figures I.2 List of Tables I.3 List of Equations I.4 List of Symbols I.5 List of abbreviations Appendix II Presentations and Publications II.1 Poster Presentations II.2 Oral Presentations II.3 Second Author Oral Presentations II.4 Publications Page 9 of 246

11 Chapter 1 General Introduction Page 10 of 246

12 1. High-Performance Liquid Chromatography Chemical analysis and purification of complex mixtures is challenging, and chromatography has been developed to address that. The technique of chromatography was originally developed by Tswett in the early 1900 s to purify coloured plant extracts. Tswett s technique is similar to modern use: a mobile phase transports the sample through a stationary phase to a detector. In Tswett s analysis, the column was open and the human eye was the detector. A detector has only limited ability to perform chemical analysis of multiple analytes simultaneously, therefore a chemical separation pre-detector, due to the column, provides assurance that the correct substance is analysed or collected. The stationary phase is a column comprised of a packed bed of solid particles, which perform a chemical separation of components in sample mixture, whereby each solute retains then elutes from the column. Chromatography advanced when in 1941 the theory of modern liquid chromatography (LC) was first described by Martin and Synge, applying plate theory from the fractional distillation used to purify petroleum extracts to describe the bands which develop in a separation (Martin, Synge 1941). In the 1970 s, Huber, Kirkland and Horvath introduced the principle of using small particles as column packing, and the technique progressed to be called high pressure liquid chromatography or highperformance liquid chromatography (HPLC). In HPLC, the column is filled with a sorbent, which is typically silica-based due to its high thermal mechanical stability, chemical resistance at moderate ph (3-7) (Berthod 1991) although alternative sorbents are also used (e.g. porous graphitic carbon). The sorbent is typically a packed bed of porous particles (diameter 1 50µm) where the stationary phase is the pore surface within the particles themselves. Since around the 1980 s, the stationary phase has been chemically bonded to Page 11 of 246

13 the pore surface, most commonly as a Reversed Phase Liquid Chromatography (RPLC) stationary phase, or unbonded as a bare silica. In RPLC the stationary phase is a hydrocarbon chain attached to a triethoxysilyl moiety, which bonds to the silica surface via a condensation reaction to give a surface with the ligand bonded to it (Fig. 1.1); the mobile phase is highly aqueous which favours retention of hydrophobic solutes into the stationary phase pores. Figure 1.1 : Octadecylsilyl ligand with isopropyl protection bonded to silica stationary phase In the earlier stages of HPLC, the stationary phase was prepared in-house, but this leads to inherent variability of performance, due to a variety of factors which require strict control e.g. ligand density on the surface, packing of the particles. Packing is itself a challenging Page 12 of 246

14 process to control, and a study by Kirkland summarises the journey from packing as a dark art to a scientific process (Kirkland et al. 2006). Modern HPLC columns are purchased with the stationary phase prepared and pre-packed. The coupling of ultraviolet absorbance (UV) detectors in the 1960 s (Kirkland 1968) and mass spectrometers (MS) in the 1970 s (Niessen 2003) to HPLC systems made these techniques powerful with scope for automation. HPLC is near-ubiquitously the technique used to measure and attain acceptable purity of non-volatile substances (Espada et al. 2008, Korfmacher 2005). Generic methods allow the application of a relatively small set of analytical methods to a wide variety of compound structures. For example the pharmaceutical company GlaxoSmithKline quality control tested a library of >700,000 compounds using a single RPLC method (Lane et al. 2006). This is very attractive industrially, as the alternative is method development for each compound of interest which can be timeconsuming. The ubiquitous RPLC is used to a high degree of sophistication in generic methods as part of Open access (OA, walk-up ) in drug development (Mallis et al. 2002). The user prepares a sample and follows on-screen instructions, the OA method analyses it and s them the result. Major pharmaceutical companies have invested in this approach (Mallis et al. 2002, Korfmacher 2005, Espada et al. 2008, Dunn April 2013). 2. HPLC separation In a HPLC separation, the sample is injected into the flow of mobile phase. The amount of mobile phase needed to do that depends on retentivity of the solute in the column and how much mobile phase is delivered by the pump during the separation. Vm = t0f (1.1) Page 13 of 246

15 Vr = Vm(1 + k) (1.2) The volume of mobile phase in one column volume (Vm) is a function of the time taken by an unretained species to pass through the column (t0) and the volumetric flow rate (F) (Equation 1.1). The volume of mobile phase required to elute a peak is the retention volume (Vr) (Equation 1.2); this is proportional to the flow rate, which can vary between separations. Guiochon commented that absolute retention times are poorly reproducible and retention factors are the favourable measure of solute retentivity (Guiochon et al. 2013). k = tr t0 t0 (1.3) The retention factor (k) is a dimensionless measure of solute retention, describing the retention of a solute relative to the passage of an unretained volume of mobile phase through the column (Equation 1.3). Samples typically contain greater than one chemical component; therefore retention must be different for each component to achieve separate peaks and allow the detector to interpret a single signal at a time. α = k2 k1 (1.4) The relative retention of two peaks is the selectivity factor (α), described by the relative retention of two closely-eluting peaks (Equation 1.4). This must be greater than one to achieve separation, and as an approximate rule of thumb, good separations are obtained for selectivity factors above 1.5. N0.5 = 5.54 ( tr W0.5 )2 (1.5) Page 14 of 246

16 To separate complex mixtures, each peak must be sufficiently narrow. Ideally peak widths would be infinitesimally small, however band broadening occurs both outside the column due to dead volumes in the instrument, e.g. tubing, and inside the column due to mass transfer and diffusion. To understand band broadening, the concept of theoretical plates was derived by Martin and Synge (Martin et al. 1941), where each solute band is analogous to a plate used to capture distillate in fractional distillation of petroleum components. Peak efficiency (N0.5) is the peak width at half-height (W0.5) relative to the retention time (Equation 1.5), units are number of theoretical plates per column. N = H L (1.6) The height equivalent to one theoretical plate is the efficiency divided by the column length (L) (Equation 1.6). State of the art Ultra High Performance Liquid Chromatography (UHPLC) systems are specially designed to minimise the extra-column band broadening due to e.g. excess tubing length. u = L t 0 (1.7) H = A + B + Cu (1.8) u Intra-column band broadening, represented by H, can be described by three processes: axial diffusion (A), longitudinal diffusion (B) and mass transfer (C), which are a function of the mobile phase velocity (u, equation 1.7) as described by the theory of J J Van Deemter (van Deemter et al. 1956) (Equation 1.8). The A term is supposedly unaffected by average mobile phase velocity (1.7) and can be used as a measure of packing quality. The B term dominates Page 15 of 246

17 at low flow rates when (B/u) is large and the solute band is allowed excessive time to diffuse along the column bed. The C term is dominant at high flow rates when (B/u) is small. To achieve good efficiency in analysis, plate heights (H) around 2-20 µm are required, which is slightly larger than the diameter of common stationary phase particles in those columns (1-5 µm). h = H d p (1.9) h = a + b + cu (1.10) u Particle diameters vary between columns, therefore in kinetic studies reduced plate height (h) is considered (1.9), which corrects for the particle size to give (1.10). An ongoing objective of column manufacturers is to produce reduced plate heights below around 1, corresponding to a solute band equilibrating within the diameter of a single particle (1.9). P = 2500LηF d p 2 d c 2 (1.11) Smaller H values can be achieved using smaller particles, as the shorter distance into and out of the particles allows for better mass transfer (1.8). However the system backpressure affected by particle size (1.11), column dimensions (length L and diameter dc), mobile phase viscosity (η) and flow rate (F). Modern UHPLC systems are designed to cope with high backpressures (around 1000 bar) when small particles (dp < 2 µm), narrow columns (2.1 mm i.d.) and high flow rates are used for fast analysis on short columns (L 5cm). In generic methods, bespoke method development is discouraged in favour of using optimal conditions for the majority of analyses. Therefore flow rate is kept constant and this project Page 16 of 246

18 hasn t focused on kinetic investigations involving varied flow rate. It is attractive industrially to use an analytical method which scales up to purify compounds without the need for further method development. Therefore this project focuses on HPLC methods that can be scaled up to preparative systems for purification. Preparative separations require the use of much higher flow rates (see 9 below) and therefore to avoid excessive backpressures, very small particles and high analytical flow rates are avoided in this project since these both contribute to high system backpressures (1.11). For a detailed discussion of kinetics in hydrophilic interaction chromatography, studies by Heaton (Heaton et al. 2014a, 2014c), McCalley (McCalley 2007) and Gritti/Guiochon (Gritti et al. 2013c, 2015) give this topic thorough consideration. Rs = ( 1 ) [ k ] (α 1) N (1.12) 4 1+k If there is sufficient selectivity (α) between peaks, efficiency has limited effect on separation power. It can be shown using the equation for chromatographic resolution (Rs; Equation 1.12) that Rs is optimal at moderate retention (1 k 5), good spacing (α 1.5) and high efficiency (N 10,000 or above). log k = log EB + η H σs* + β A + α B + κ C (1.13) The basis of retention in RPLC is the interaction of the solute with the stationary phase. A central tenet of that is solute partition from a mostly aqueous mobile phase into a layer of octadecylsilyl (ODS, also known as C-18) ligands. This is ideal for non-polar solutes with low affinity for aqueous media and high affinity for the hydrophobic environment inside the column pores. Different interactions between solute and stationary phase are possible and Carr, Snyder et al. described those by the Hydrophobic-Subtraction model (Equation 1.13), Page 17 of 246

19 where the effect of each interaction on retention is considered (Carr et al. 2011, 2015, Marchand et al. 2011). Retention (log k) is a function of: partitioning from the mobile to stationary phase (represented by the log of retention of a neutral solute ethylbenzene (EB)), hydrophobic interactions (η H), steric interactions (σs*), hydrogen bonding of a basic solute to an acidic stationary-phase group (β A), hydrogen bonding of an acidic solute to a basic stationary-phase group (α B) and ion-exchange between an ionic solute and a chargebearing column (κ C). However highly polar solutes are either not retained by RPLC or resolved poorly (McCalley 2010a) due to a low affinity for hydrophobic C18 stationary phase relative to a highly aqueous mobile phase. Therefore alternative separation modes have been considered. 3. Polarity and Hydrophilicity To establish if a solute is hydrophilic and unlikely to retain by RPLC, it is possible to measure a solutes hydrophilicity using two immiscible solvent phases, normally water and an organic solvent such as n-octanol. X(aq) X(org) (1.14) A hydrophilic species will partition into the aqueous phase (X(aq)) and a hydrophobic species will partition into the non-aqueous phase (X(org)) (Equation 1.14). P = C(Xaq) C(X org) (1.15) Page 18 of 246

20 The partition coefficient of this process (P) is calculated by measuring the concentration of the solute in the aqueous [C(aq)] and organic phases [C(org)] (Equation 1.15), for a solute in its neutral form which can be achieved by adjusting the ph. log P = log ( CXaq ) log (CXorg) (1.16) However P can vary over several orders of magnitude depending on the solute. Log P (Equation 1.16) is a simple value that is increasingly positive for hydrophobic solutes that partition into the organic solvent and increasingly negative for hydrophilic solutes that partition into the organic solvent. However for complex mixtures each solute may be neutral or charged and measuring P is difficult. HA = H 3 O + + A - (1.17) Ka = [H3O+ ][A - ] [HA] (1.18) ph = pka + log [A ] [HA] (1.19) To calculate the acidity or basicity of a solution and calculate the solute charge state requires ph and pka calculations. For the dissociation of an acid (HA) to hydroxonium (H3O + ) and its anion (A - ) in the presence of water (Equation 1.17), the dissociation constant (Ka) is described by equation 1.18, the concentration of each species shown in square brackets. The negative of the log hydroxonium concentration is equivalent to the ph, and logka is the pka, which relate as shown in equation log D = log P + log [ 1+10 pka ph] (1.20) 1 Page 19 of 246

21 The distribution coefficient (D) calculates the partitioning of a solute in its native form whether ionised or neutral, and log D this takes into account the solute pka, solution ph and log P (equation 1.20). The log D value positive for hydrophobic species that partition into the non-aqueous portion or negative for hydrophilic species that partition into the aqueous portion (equation 1.20). Log D correlates well with hydrophobic retention in RPLC, where a more-positive log D value corresponds to stronger retention on those columns (Poole 2009). 4. HPLC of polar pharmaceuticals RPLC is highly productive using modern columns and modern systems that can cope with high backpressures from fast-flowing mobile phase through small particles on narrow columns, however its application is limited to hydrophobic solutes. Fragment-based drug discovery uses small molecules as potential new drugs, described in a seminal paper by Jencks in 1981 (Jencks 1981). This strategy focuses on optimising interactions between chemical species and proteins, and these small molecules (<300 Da) can then be chemically modified to improve physico-chemical properties such as bioavailability. This strategy has since been adopted across the pharmaceutical industry, and was described by Scott et al. as firmly established in drug discovery (Scott et al. 2012). A problem with this strategy is these small molecules can be hydrophilic, with weakly basic or zwitterionic chemical functional groups providing potential biological activity and capacity for formation of C-X bonds in further synthesis (Scott et al. 2012, Jencks 1981). This poses a problem for laboratories that synthesise drug-fragments: reliable purity measurements are essential for their quality control prior to high-throughput screening for biological activity (Espada et al. 2008). The hydrophilicity of Molecular building blocks used in fragment-based drug design Page 20 of 246

22 is a serious problem since well-established RPLC requires a solute to be hydrophobic in order to retain on those columns. Polar compounds aren t retained by RPLC and chromatographic separation can be impossible by this somewhat traditional method. Some alternative variations of RPLC have been developed to manipulate the hydrophobicity of the solute and gain retention on those columns. Low ph mobile phase can be prepared by adding a weak or strong acid such as formic (FA) or trifluoroacetic acid (TFA), respectively. When an acidic solute is deprotonated it is negatively charged and more hydrophilic. Conversely, adding a strong acid to this will protonate, thus neutralise, the acid, and it is more hydrophobic. Therefore FA and TFA are used to retain some acids by RPLC. A similar strategy is used for basic solutes: the protonated form of a base is positively-charged and more-hydrophilic; the unprotonated form is neutral and more-hydrophobic. Thus adding a strong base to RPLC mobile phase raises the ph, neutralises basic species and enhances hydrophobic retention on those columns (McCalley 2004, Davies et al. 2008). A potential flaw in the high ph technique is the liability of silica to dissolve through hydrolysis by hydroxyl ions (OH - ). This can be overcome using hybrid silica, which substitutes ethylene bridges for siloxane bonds between silanols in the underlying structure (see poi. This hybrid silica is somewhat resistant to high ph (see point 2 on p. 24). These strategies can be successful for analysis of simple acids and bases by RPLC, but hydrophilic neutral species and zwitterions are unsuitable for high and low ph RPLC, since their pka s don t allow for enhanced hydrophobicity at extremes of ph. Page 21 of 246

23 5. Ion pair chromatography Addition of Ion-Pairing (IP) agents such as trifluoroacetic acid (TFA) also facilitate retention of charged polar species. A drawback of IP is a reduction of detector sensitivity, especially in Electrospray Mass Spectrometry (ESI-MS) where IP agents can suppress analyte ionisation (Heaton et al. 2011). Additionally, it is unclear what the precise retention mechanism is in IP, for example Dai et al. reported for basic solutes that approximately 3% of molecules associate with the ion pair agent TFA in aqueous solution (Dai et al. 2005). 6. Hydrophilic Interaction Chromatography Polar solutes are hydrophilic and require separation on stationary phases that can attract such species. Hydrophilic Interaction Chromatography (HILIC) is a variant of HPLC has been used to separate sugars since at least the 1970 s and in 1990 Alpert coined the name HILIC, which coincided with the release of columns specifically designed to use this separation mode to analyse e.g. phosphorylated amino acids and peptides (Alpert 1990). In HILIC the stationary phase is either bare silica or a bonded polar ligand. The mobile phase has high organic solvent content (>70% ACN) with small water content and buffer. Water forms a stagnant layer on the stationary phase; this allows solutes to partition between a hydrophobic mobile phase and a hydrophilic stationary phase (Figure 1.2). Page 22 of 246

24 Figure 1.2. Simple scheme of HILIC retention with neutral (X), basic (X+) and acidic (Y-) solutes The specific interactions between solute and stationary phase have been debated in the literature (Irgum et al. 2006, 2011, Kawachi et al. 2011, McCalley et al. 2010, 2013, 2014, Gritti 2013c, Guo et al. 2005, Laemmerhofer et al. 2008, Bicker et al. 2008). That discussion primarily discussed whether or not partitioning is the dominant retention mechanism in HILIC, as proposed by Alpert (Alpert 1990). A 2006 review by Irgum of HILIC literature was inconclusive, with some of the authors covered suggesting surface-solute interactions, e.g. on Amino columns (Irgum et al. 2006). It was suggested by LammerHofer Lindner and Bicker that the retention mechanism is complex, with contributions from partitioning, ionexchange and hydrogen-bonding (Laemmerhofer et al. 2008, Bicker et al. 2008). McCalley demonstrated that ion-exchange can be mediated by the buffer salt concentration, with high buffer concentration shielding the solute from stationary phase charges (McCalley 2010b). In that study it was noted by the author that ion-exchange differs greatly between HILIC columns (McCalley 2010b). Studies by Irgum, Ikegami and McCalley attempted to categorise HILIC columns according to their retention behaviour (Dinh et al. 2011, Kawachi et al. 2011, Kumar et al. 2013). There Page 23 of 246

25 was good agreement that HILIC columns can be described as four broad categories (ligand chemistry described in brackets, approximate structures shown): 1. Cation-exchangers (e.g. Bare Silica) (e.g. Silica Hydride) 2. Neutral polar bonded ligand (e.g. Amide, BEH Amide) Spacer 3. Zwitterionic polar bonded bonded ligand (e.g. ZIC- HILIC) Page 24 of 246

26 4. Anion Exchangers (e.g. Amino) The Tanaka group were critical of ion-exchange in HILIC, commenting that reducing ionexchange interactions is important to obtain better column efficiency in HILIC, with apparent peak tailing when this interaction is employed (Kawachi et al. 2011). However Kumar et al. observed higher theoretical plate numbers on bare silica than bonded-phase columns (Kumar et al. 2013). Ikegami noted that retention differences could be observed between neutral nucleosides and their corresponding nucleobase (e.g. k (uridine)/ k (uracil) reported as 1.81 on an Amide ligand-bonded silica column). The authors commented that the number of hydroxyl groups on the ribose moiety has a great influence on the retention of nucleosides on HILIC columns, suggesting uridine series are suitable probes for HILIC studies (Kawachi et al. 2011). Exceptionally high retention of the nucleoside cytidine on Amide and ZIC-HILIC columns was reported by Kumar et al. and those authors suggested this Page 25 of 246

27 solute can hydrogen-bond with the stationary phase (Kumar et al. 2013). An alternative type of silica was developed around 1991 and has been developed by Pesek as type C silica. On these columns, acidic silanols groups are replaced by silica hydride groups and there ought to be virtually no exposed silanols (Yang et al. 2013). Applications using type C silica use formic acid as a mobile phase additive, which is reasonable given the manufacturer s claim of an inert stationary phase. However, a report from Watson concluded that type C silica behaves remarkably similar to bare silica (Bawazeer et al. 2012) and this phase has shown poor peak shapes in some literature when using Formic Acid as buffer (Yang et al. 2013). It is unclear if type C silica offers alternate selectivity compared to bare silica and therefore that was considered in this study. Acetone has been used as a mobile phase organic modifier in place of acetonitrile by the Haddad group (Hutchinson et al. 2012) and by Heaton for its use in MS (Heaton et al. 2011), but it is unclear how retention compares in acetone to the more typical acetonitrile. Some more recent fundamental studies into HILIC focused on mass transfer (Gritti et al. 2013b, 2013c, Heaton et al. 2014a). Solute mass transfer occurs in the mobile phase and the stationary phase (1.18). The contribution of each can be determined in kinetic studies, where theoretical plate heights (1.6) are plotted over a range of average mobile phase velocities and the results fitted to equations such as van Deemter equations (1.8) or (1.10). Mass transfer in the mobile phase is perhaps expected to be high in HILIC due to the low viscosity of organic solvent-rich mobile phase providing relatively free movement of solutes. A 2010 study reported somewhat high mobile phase mass transfer in HILIC using a bare silica column compared to RPLC on a C18 column (McCalley 2010b). Gritti and Guiochon also reported higher van deemter B terms in HILIC compared to RPLC (Gritti et al. 2013c). Page 26 of 246

28 However when those authors used peak parking to monitor solute movement in the stationary phase, where solute band diverted is diverted to a second column with no mobile phase flow and allowed to diffuse along the column bed, they found diffusivity was low, in contrast to high diffusivity in the mobile phase. The authors attributed that to a relatively high microviscosity of the water layer held to the stationary phase (Gritti et al. 2013c). Heaton et al. measured diffusion of hydrophilic species on comparable columns with matched retention factors and observed a similar effect, suggesting that adsorption via possible hydrogen bonding between solute and stationary phase surface can contribute to retention (Heaton et al. 2014a). HILIC is compatible with polar compounds, although is yet to be incorporated into a generic method scheme for the analysis of polar pharmaceuticals. The retention mechanism in HILIC is complex and poorly understood, therefore research into the HILIC retention mechanism, with a focus on selectivity and peak efficiency, should lead to a polar tool box of generic methods. The role of buffers in HILIC is unclear, and requires investigation on a range of modern HILIC columns. Detection by MS is less effective in the buffers typically used for HILIC methods, due to analyte signal suppression (Kostiainen et al. 2009, Mallet et al. 2004, Law et al. 2000). An objective of this project is therefore to investigate if formic acid can be used as a buffer in HILIC generic methods, as opposed to buffers. 7. HPLC detection Batches of drug leads are often products of one-off syntheses, containing impurities which are unknown and standards are thus unavailable. Detection is typically by UV which is cheap and simple to operate or Mass Spectrometry which gives assurance of compound identity Page 27 of 246

29 via mass/charge data. Impurity compounds may not contain chromophores thus ultraviolet wavelength (UV) detectors are blind to them. Using MS and UV, small peaks are not meaningful in the absence of reference standards and are not necessarily impurities. Therefore universal detectors, such as the charged aerosol detector (CAD) are a desirable component of a polar tool box. However CAD is a new technology and although operation is very straightforward (Vehovec et al. 2010), it is poorly understood. 8. Charged Aerosol Detection A prototype detector was built by Dixon and Peterson in 2002 (Dixon, et al. 2002) and since then CAD has been developed for use in HPLC. Around 100 publications to date have focused on CAD (e.g. (Cohen et al. 2012, Gamache et al. 2005, Web of Science search topic Chromatograph* AND TITLE Charged Aerosol* ), but only a handful explored the fundamental properties of this relatively novel detector (Dixon et al. 2002, Gamache et al. 2005, Hutchinson et al. 2010, Hutchinson et al. 2012, Khandagale et al. 2013, Vervoort et al. 2008). Figure 1.3 shows a simplified schematic of the CAD. The CAD produces aerosol particles (Steps 1-2 in Figure 1.3) and positively-charged nitrogen gas. These mix such that the aerosol particles acquire positive surface charges (Step 3 in Figure 1.3), and are then transported to an electrometer which converts their charge to an electrical signal (Step 4 in Figure 1.3) via in-built hardware. In contrast to UV detectors, the CAD does not require a solute to contain a chromophore due to it forming physical particles of whatever solute is present. In MS, molecular ions are formed which is in contrast to CAD which forms charged particles. Although some aspects of CAD operation are already understood, it is not clear how it responds to semi-volatile and volatile solutes, which ought to be incapable of forming Page 28 of 246

30 aerosol particles. Other detectors which depend on the formation of aerosol particles, such as evaporative light scattering detection (ELSD) as developed by Charlesworth in 1970 (Charlesworth 1978), suffer from complex relationships between solute concentration and detector response (Guiochon et al. 1988). ELSD, which measures the light scattering of a laser when the aerosol particles cross the beam, is thought to change in detection mechanism with increasing size of particle. CAD is somewhat more straightforward but nonetheless also depends on aerosol particle formation. It is possible there is some commonality between the ELSD and CAD theory insofar as particle formation is concerned. Thus an empirical relationship between solute concentration and detector response might be achievable. A combination of universal detection and a universal response to solute concentration suggest CAD has potential as a HPLC detector in generic methods. Therefore an objective of this project was to evaluate the performance of CAD, in particular with the use of HILIC separations. Figure 1.3. The Charged Aerosol Detector Page 29 of 246

31 9. Purification of polar pharmaceuticals Purification by preparative HPLC uses wider-bore columns compared to analytical separations ( 10 mm i.d. preparative, 4.6 mm i.d. analytical) to hold sufficient stationary phase so that larger samples can loaded onto the column. Scale up factor = dcprep2 dc Analytical 2 (1.21) When scaling up a separation, to maintain the same average mobile phase velocity the flow rate is scaled up in proportion to the ratio of the squared column diameter (1.21). The injection volume is scaled up by the same factor (1.21) to maximise the loading of sample. Preparative HPLC commonly employs sample loads far above the column capacity, and separation performance is degraded as a result of shifts in retention and broad peaks with low efficiency. Purification studies using HILIC are scarce in the scientific literature, although this is a necessary application of the technique. McCalley reported in 2007 that bare silica HILIC column(s) have capacity around ten times higher than RPLC for strong bases (McCalley 2007), which are particularly problematic in RPLC (McCalley 2010a). Gritti and Guiochon studied the overloading of strong bases propranolol and amitriptyline hydrochloride using a bridged ethylene hybrid (BEH) silica in HILIC (Gritti et al. 2015), reporting similar improvements over RPLC. This study used a charged surface hybrid (CSH)-C18 RPLC column, which contained positive charges to control solute repulsion, which is thought to be responsible for the tailing overload of charged bases even at low solute concentration (Gritti et al. 2015). Bonded phase columns are available with diverse chemistries in HILIC, which can provide substantial changes in selectivity (Kumar et al. 2013, Kawachi et al. 2011, Dinh Page 30 of 246

32 et al. 2011). However none of these columns have featured in HILIC purification or loadability studies. 10. Focused Gradient Liquid Chromatography and At- Column Dilution The majority of fundamental studies into HILIC have used isocratic conditions, whereby the mobile phase composition is held constant throughout the separation. This simplifies the methodology, however there are practical benefits from changing the mobile phase during the separation (Snyder et al. 2010). In RPLC, applying a gradient of mobile phase organic solvent content is common practise, starting from mostly-aqueous mobile phase to a more organic-rich mobile phase. The benefits from this are reduced run time, as the increased organic solvent content elutes strongly-retained solutes from the column. Additionally, peak shape is improved using a solvent gradient: once eluted the solute travels solely into the mobile phase and interactions with the stationary phase are minimal, therefore band broadening is less pronounced. Solvent gradients have been applied to HILIC separations (Karatapanis et al. 2009, Periat et al. 2013a, Tyteca et al. 2014), however in contrast to RPLC the starting conditions are an organic-rich mobile phase changing to a more-aqueous mobile phase. Focused gradients can be used to expand areas of the chromatogram and are used in day-to-day preparative work in industry, although publications on this technique are limited to application notes, e.g. (Tei et al. 2013). A recent study of optimisation of relevant parameters in preparative separations by Forssén and Fornstedt found that selectivity (α) was the most important factor to maximise productivity (Forssén et al. 2014). Focused methods are designed to improve the spacing between the peak of interest and the nearesteluting species, although focused gradients have not been reported to been applied using Page 31 of 246

33 HILIC. Therefore that strategy was employed in this project to HILIC-prep separations. Atcolumn dilution was described by Neue to aid loading of poorly-soluble compounds (Neue et al. 2003). In this technique, the sample is introduced onto the column slowly via a second pump, which is diluted at the column head by a second flow of weakly-eluting mobile phase mixed in to the sample flow using a T-piece (Fig. 1.4). This was employed in this project to further enhance preparative performance. Figure 1.4. At Column Dilution Page 32 of 246

34 11. Objectives The principle interest of this project is the analysis and purification of hydrophilic drug-like solutes and polar building block molecules. The present study evaluates HILIC in a generic method setup. Choice of some basic parameters is necessary before HILIC methods can be implemented in a generic setup, namely the following. 1. Suitable mobile phase buffer A detailed study of mobile phase buffers including salt and simple acid buffers was necessary. Previous studies by Watson et al. have suggested simple acid buffers may not be suitable on type C silica phases (Bawazeer et al. 2012), thought to contain Si-H bonds as ligands (Yang et al. 2013). A study using a bare silica HILIC column suggested formic acid may not be suitable for basic solutes (McCalley 2007), but this has not been evaluated for alternative bonded-phase columns. 2. Suitable stationary phase Studies by Irgum, Ikegami and Kumar have shown differences in the retentivity of HILIC stationary phases (Dinh et al. 2011, Kawachi et al. 2011, Kumar et al. 2013). Based on this, and the categories described in (6), a bare silica and BEH Amide column were chosen for generic HILIC use, as these give appreciably alternate selectivity (Kumar et al. 2013) and the BEH Amide has extra stability resulting from the bridged ethylene hybrid silica with potential for future alternative ph use (McCalley 2015). Page 33 of 246

35 3. Universal detection Charged aerosol detection is supposedly a universal detector, responding to any nonvolatile solute, with uniform response, independent of solute chemistry. The calibration curves of solute concentration vs. detector response ought to be simpler than established universal detectors such as ELSD. However these factors have not been evaluated with a sufficiently broad range of solute chemistries. Additionally, the organic-rich solvent used as HILIC mobile phase ought to give excellent detector response due to facile desolvation, as reported for MS elsewhere (Periat et al. 2013b). This has not been evaluated by HPLC for the CAD. It was therefore an objective of this project to elucidate the effect of various parameters and conditions on CAD response with a view to describe optimal use of this relatively novel detector. 4. Viability of HILIC purification. It is industrially attractive to scale-up analytical methods directly to larger-bore preparative columns for purifications, and although this has been attempted for mixed-mode and aqueous normal phase methods, HILIC can be operated at lower mobile phase salt concentration than these modes (e.g. 5mM cf. 20mM) which simplifies work-up as the salt is removed to produce a pure product (Underwood May 2014). It was therefore an objective of this project to develop some generic HILIC analytical methods suitable for polar pharmaceuticals and apply those to purification in a proof-of-concept study. 5. Suitable sample diluent A basic understanding of hydrophilic compound solubility in HILIC mobile phases has not been established in the literature. One study by Guillarme et al. reported the water content Page 34 of 246

36 of diluents used in sample preparation must be kept to a minimum for HILIC separations (Ruta et al. 2010) and alternative solvents may be possible diluents for HILIC (Ruta et al. 2010). Further work in this area is crucial if HILIC can be employed to purify polar solutes on the scale required by the pharmaceutical industry. Page 35 of 246

37 Chapter 2 General Experimental Page 36 of 246

38 1. Instrumentation a. HILIC buffer experiments These were performed with a 1290 binary high pressure mixing UHPLC instrument (Agilent, Waldbronn, Germany) with Chemstation, photodiode array UV detector (0.6 µl flow cell) and 5 µl injections. b. Charged Aerosol Detector experiments These were performed with a Thermo UltiMate 3000 Rapid Separation Liquid Chromatography system. This was comprised of a quaternary pump, diode array detector (DAD) and either a Corona Ultra or Corona Veo CAD, with Chromeleon 7.2 software (Thermo, Germering, Germany). The CAD is a destructive detector, therefore the DAD and CAD detectors were connected in series in some experiments, with flow first through the DAD. Thermo Viper tubing (0.13 mm ID) was used as connection tubing. Data collection rates were 100 Hz for both DAD and CAD, due to narrow peak widths (typically 1 s at half height in flow injection analysis (FIA)). The Corona Ultra nebuliser (cross flow design similar to that used in atomic absorption spectrometry) was controlled at 22 C with the evaporator tube at ambient temperature, while the Veo (concentric flow design similar to those used in mass spectrometry) nebuliser was at ambient temperature and the evaporator tube set to 30 C. The Veo had a power function (PF) designed to linearise data, which was set to either 0.67 (this simulates off ), 1.00 (the default) or 1.2 (optimised setting using experimental data, see below). Experiments on acetone as a HILIC mobile phase and dimethylsulfoxide (DMSO) as a diluent on analytical columns were also performed on this Thermo system. Page 37 of 246

39 c. HILIC Generic and focused method development These were performed at GlaxoSmithKline laboratories, using an Agilent 1100 system (Agilent, Waldbronn, Germany) with Chemstation, binary pump, UV Diode Array Detector (DAD) and 1 µl injections. d. HILIC prep experiments These were performed at GlaxoSmithKline laboratories, using a Waters prep system with Masslynx, quaternary pump, and automated fraction collection, QDa mass spectrometer with an electrospray interface, UV diode array detector and UV post-fraction detector. Fractionation was directed by the MS, which was set to sufficiently high sensitivity such that fractions were discarded to waste in this proof-of-concept study. 2. Conditions a. Injection Chromatographic peak shape can be sensitive to the injection volume used to introduce the sample. Dolan advised this be limited to around 15% of the peak volume (Dolan 2014) with a rough rule of thumb to keep injection volumes below 16uL for the analytical column and particle dimensions used in these studies (Dolan 2014). Above this, volume overload can occur, which reduces peak efficiency thus reduces resolution (Rs). Injection volumes were kept well below this, as the UV detectors offered sufficient sensitivity at 5 µl injections. The CAD experiments used low injection volumes of typically 1 µl, as an objective of those studies was to establish the detection limits of this relatively novel detector. This injection Page 38 of 246

40 volume (1 µl) was also found suitable for flow injection analysis (Chapter 4, 5). Preparative experiments used custom injection volumes as described in Chapter Chemicals and reagents All test solutes, and rubidium nitrate were obtained from Sigma-Aldrich (Poole, U.K.). Acetonitrile (ACN, far UV grade), ammonium formate (AF) and orthophosphoric acid (PA) were obtained from Fisher (Loughborough U.K.). AF buffers were prepared by adjusting aqueous solutions to ph 3.0 with formic acid such that the over-all concentration of AF in the mobile phase after organic solvent addition was 5 mm. The ph values of the mobile phase quoted are those either in the aqueous portion of the buffer (w w ph) or alternatively as measured in the organic-aqueous combination with the electrode calibrated in aqueous buffers (w s ph). Standards for HILIC buffer experiments were prepared at a concentration of 50 mg/l and made up in the exact mobile phase. For CAD experiments these were prepared at a concentration of typically 10,000 mg / L in ACN-water with 0.1% FA (v/v), then diluted with exact mobile phase to the required concentration. For HILIC generic method development, a combined standard of eight probe solutes was prepared at a concentration of 0.5 mg / ml in ACN-water with 0.1% FA (v/v). For HILIC prep experiments, custom diluents and concentrations were used as described in Chapter 6. Page 39 of 246

41 4. Probe solutes To represent a variety of polarities, hydrophilicities and charge states, a selection of neutral, acid, basic and zwitterionic compounds were used as test probes. Each results chapter describes the solutes chosen for that particular study. Page 40 of 246

42 Chapter 3 Comparison of peak shape in hydrophilic interaction chromatography using acidic buffers and simple acid solutions

43 Abstract The retention and peak shape of neutral, basic and acidic solutes was studied on hydrophilic interaction chromatography (HILIC) stationary phases that showed both strong and weak ionic retention characteristics, using aqueous acetonitrile mobile phases containing either formic acid (FA), ammonium formate (AF) or phosphoric acid (PA). The effect of organic solvent concentration on the results was also studied. Peak shape was good for neutrals under most mobile phase conditions. However, peak shapes for ionised solutes, particularly for basic compounds, were considerably worse in FA than AF. Even neutral compounds showed deterioration in performance with FA when the mobile phase water concentration was reduced. The poor performance in FA cannot be entirely attributed to the negative impact of ionic retention on ionised silanols on the underlying silica base materials, as results using PA at lower ph (where their ionisation is suppressed) were inferior to those in AF. Besides the moderating influence of the salt cation on ionic retention, it is likely that buffers improve peak shape due to the increased ionic strength of the mobile phase and its impact on the formation of the water layer on the column surface. Page 42 of 246

44 1. Introduction Hydrophilic interaction chromatography (HILIC) is rapidly establishing itself as a complementary technique to reversed-phase separations (RP), particularly for polar and/or ionised compounds that are poorly retained using the latter method. It is a technique wellsuited to the analysis of pharmaceuticals and compounds of biomedical significance (Olsen 2001, Periat et al. 2013b, Zhou et al. 2008). The stationary phase in HILIC is typically bare silica, or polar groups bonded to a silica or an organic polymer matrix (McCalley 2010b, Kawachi et al. 2011, Hemstrom et al. 2006). The hydro-organic mobile phase is similar to that used in RP, except typically employs much higher concentrations of acetonitrile (>70%). There is appreciable overlap in the applicability of these two techniques to compounds of moderate hydrophilicity, particularly for basic compounds. These can be retained by ionic interactions which occur on all silica-based phases as well as by hydrophilic interactions (McCalley 2010b, 2013, Kawachi et al. 2011, Hemstrom et al. 2006). Hydrophilic interactions are likely to result from a combination of solute partition between a water layer held on the surface of the column and the bulk mobile phase, and by adsorption onto polar groups that may be partially deactivated by the presence of the water layer (Hemstrom et al. 2006). HILIC separations are usually performed in ACN-water mobile phases containing additives or buffer components, particularly when the analysed solutes are ionogenic. The buffer serves to control the ionisation of the stationary phase surface groups and silanols in silica-based phases, as well as the ionisation of the solute. The choice of buffers for HILIC is limited to those that have sufficient solubility in high concentrations of ACN. Typically, ammonium acetate or ammonium formate (AF) is used; these salts have the additional advantage that they are volatile and thus compatible with nebuliser-based detectors e.g. electrospray Page 43 of 246

45 ionisation mass spectrometry. However, use of buffers can cause depression of the electrospray signal that increases with concentration over the typical range (5 50 mm) employed (Kostiainen et al. 2009, Mallet et al. 2004, Law et al. 2000). Even at the 5 mm level, it was shown that AF can cause greater signal suppression for acidic and basic pharmaceuticals compared with the use of simple acidic solutions of 0.1% formic acid (FA), which are commonly used. An added advantage of these acid solutions is that they are easier to prepare than mobile phases containing buffers. Nevertheless, it has been shown that ACN-water mixtures containing formic acid alone can give rise to poor peak shape in HILIC for acidic and basic solutes, whereas good peak shapes were obtained with AF buffers (McCalley 2007). However, these studies were performed solely on a bare silica column. It is possible that the strong ionic interactions with ionised silanols on this type of phase are contributory to this poor peak shape with FA, and that buffers are unnecessary with other types of HILIC columns (Kumar et al. 2013). For example, bonded phase (e.g. with amide ligands) materials prepared on inorganic organic hybrid silicas show much reduced ionic interactions. Furthermore, silica hydride materials (Type C silica) are available for HILIC-type separations. It is claimed that this new type of stationary phase has significant differences in terms of chemical structure to traditional silicas, which are mainly populated with polar silanol groups. In contrast, Type C silica apparently has surface silicon-hydride groups (Pesek et al. 2008, Boysen et al. 2011). The term aqueous normal phase (ANP) has been suggested to describe separations on this type of silica phase to distinguish them from classical HILIC separations. Nevertheless, ANP is also a term more generally used as an alternative to HILIC for classical separations, reflecting the possibility that adsorption is at least a contributory mechanism along with partition to the overall retention mechanism. It could be supposed that these Type C stationary phases would contribute considerably less Page 44 of 246

46 ionic inter-actions, so the use of buffers might be unnecessary with such phases, if ionised silanol groups were the cause of peak shape problems. Indeed, separations on these phases are often reported with ACN-water mixtures containing only 0.1% formic or acetic acids (Boysen et al. 2011, Pesek et al. 2008, Bawazeer et al. 2012) although no comment has been made in these reports concerning the lack of use of buffers, or whether their absence gave rise to any detrimental (or even beneficial) effects. The aims of this paper were to compare the use of buffers with acid solutions for acidic, basic and neutral solutes separated on a variety of stationary phases, including bare silica, amide bonded onto hybrid silica, zwitterionic and silica hydride phases. These materials are considerably different in their retention characteristics towards ionised solutes, and therefore might produce different results in the various mobile phases. In this way we hoped to gain information to assist appropriate mobile phase selection for use in HILIC and HILIC with mass spectrometric detection. This study is divided between initial work to establish a pragmatic buffer choice for ongoing work in the project (Chapters 4-6), and detailed studies to further elucidate the phenomena responsible for those results. 2. Experimental Initial experiments were peformed with an Agilent 1100 binary HPLC instrument (Agilent, Waldbronn, Germany) with Chemstation, UV variable wavelength detector and 5µL injections. All experiments were performed with a 1290 binary high pressure mixing UHPLC instrument (Agilent, Waldbronn, Germany) with Chemstation, photodiode array UV detector (0.6 µl Page 45 of 246

47 flow cell) and 5 µl injections. The columns used (all cm ID, except where stated) were Cogent Silica C (4 µm particle size, pore size 100 A, surface area 350 m 2 /g) from Microsolv (Eatontown, USA), Atlantis silica (5 µm particle size, pore Size 100 Å, surface area 360 m 2 /g) from Waters (Milford, USA), ZIC-HILIC (5 µm particle size, pore size 200 Å, surface area 140 m 2 /g) from Merck-Sequant (Umeå, Sweden) and XBridge BEH Amide (15 cm 0.46 cm, 3.5 µm particle size, pore size 140 Å, surface area 190 m 2 /g) from Waters. By replacing the column with a zero dead volume fitting, the extra-column bandspreading of the instrument was estimated to reduce column efficiency by less than 5% even for a nonretained peak on the most efficient column. Temperature was maintained at 30 C using the Agilent column compartment. Acetonitrile (far UV grade), ammonium formate and orthophosphoric acid were obtained from Fisher (Loughborough U.K.). AF buffers were prepared by adjusting aqueous solutions to ph 3.0 with formic acid such that the over-all concentration of AF in the mobile phase after organic solvent addition was 5 mm. Standards were prepared at a concentration of 50 mg/l and made up in the exact mobile phase. The ph values of the mobile phase quoted are those either in the aqueous portion of the buffer (w w ph) or alternatively as measured in the organic-aqueous combination with the electrode calibrated in aqueous buffers (w s ph). All test solutes, and rubidium nitrate were obtained from Sigma-Aldrich (Poole, U.K.). Log D and log P values were calculated as the average from three different programs: ACD version 12.0 (ACD labs, Toronto, Canada), Marvin (ChemAxon, Budapest, Hungary) and MedChem Designer (Simulations Plus, Lancaster, USA). pka and solute charge was calculated from the average estimate given by the first two calculators. For the initial studies column efficiency was measured at half-height. For the detailed studies to elucidate explanation for those data, column efficiency (N) was measured from the first and second statistical moments according to the relationship in 3.1. Page 46 of 246

48 Asymmetry factor was measured at 10% of peak height by dividing the width of the trailing edge of the peak by that of the leading edge. The columns were operated in the region of their optimum flow (1.0 ml/min for silica and hydride silica, 0.5 ml/min for zwitterionic and amide). N = M12 M 2 (3.1) Page 47 of 246

49 Fig Structures, pka, log P/D and charge at w w ph 3 and w w ph 5 for the probe solutes Page 48 of 246

50 3. Results and discussion 3.1. Buffer and solute properties. Table 3.1 indicates the ph, ionic strength and buffer capacity of the three mobile phases used, 5 mm ammonium formate (AF) adjusted to ph 3.0 with formic acid, 0.1% (v/v) formic acid (FA), and 0.1%(v/v) orthophosphoric acid (PA), if prepared in aqueous solution. Ammonium formate and formic acid are soluble in high concentrations of ACN; they are also volatile additives and thus extremely suitable for use in HILIC with mass spectrometry detection (Periat et al. 2013b). PA is an alternative acid additive used by several column manufacturers e.g. (Halo Penta-HILIC brochure 2014). It was used successfully by Mant and Hodges for the HILIC separation of peptides using a 0.2% concentration in 85% ACN, using UV detection (Mant et al. 2008). These authors sought a more hydrophilic acid additive than trifluoroacetic acid (TFA). We showed by experiment in the present study that 0.1% PA was completely soluble even in 100% ACN, with no evidence of precipitation. PA is not volatile and is thus unsuitable for use with mass spectrometry detection. However, PA was studied due to the lower w w ph and w s ph given by this relatively strong acid, and thus its better ability to suppress the ionisation of residual silanol groups. PA is also not expected to give substantial ion pair effects (see the discussion of these effects in Section 3.2). Ion pairing could lead to lower retention of ionised bases due to reduction in ionic interactions with the stationary phase and the reduced hydrophilicity of the paired species. In contrast, trifluoroacetic acid, which is a stronger acid and is more hydrophobic than PA can give quite pronounced ion pair effects (McCalley 2007), which we believed might have confounded the interpretation of the results by affecting retention times. Page 49 of 246

51 A relatively low concentration of AF was employed, as such concentrations are generally preferred when mass spectrometry is used for detection. While the buffer capacity of formic acid in water is the least of the three solutions, it is still appreciable. w s ph values(in the organic aqueous mixture) are shown for 85% ACN solutions. The choice of this measurement as opposed to w w ph in the aqueous fraction alone is not straightforward. Detailed computer modelling by Tallarek and co-workers (Melnikov et al. 2011, 2012), suggests that there is a layer exclusively of water molecules tightly bonded to the surface of bare silica; in this case the use of w w ph may be more appropriate. However, other experimental work suggests there may be significant numbers of acetonitrile molecules in the interfacial region (Rivera et al. 2013). Fig. 3.1 shows the structure, pka, charge and log D/log P values at w w ph 3.0 and 5.0 for the 12 probe solutes used, which were a mixture of neutrals, strong and weak acids and bases, and a quaternary ammonium salt. Average values of these parameters from several different calculation programs were used in order to improve the accuracy of the estimations. Although the agreement of estimates from the programs was reasonable, there was some lack of consistency between the programs due to the use of different software algorithms (Kumar et al. 2013). Page 50 of 246

52 Table 3.1 ph, molarity and buffer capacity of aqueous buffer solutions and dilute acids; w s ph measured in 85% ACN. *This was used as 0.1% of an 85% solution (14.6 mm/l). Buffer w w ph w s ph Molarity Buffer capacity (mmol/l ph) (mmol/l) 0.1% Formic Acid mM Ammonium Formate ph % Phosphoric Acid* Initial studies to establish a generic HILIC mobile phase buffer: performance of four different phases with three mobile phase buffers Initially this study was performed on an Agilent 1100 system with a view to establish a generic mobile phase buffer system for the remainder of the project. Eleven solutes were used as probe compounds, shown in Fig. 3.1 with the exception of Uracil, which was added to later parts of this work (3.3) to characterise a somewhat broader selection of hydrophilic neutral solutes. Each solute was injected individually onto the respective column in mobile phase buffered either by ammonium formate (5mM w w ph 3 with formic acid), formic acid (0.1% v/v) or phosphoric acid (0.1% v/v). Plots of column efficiency, measured at peak halfheight, are shown in Fig. 3.2a-d for each column for the three buffer systems. Peak Page 51 of 246

53 asymmetry was measured at 10% peak height (As0.1). Peak shape in terms of efficiency and symmetry was superior in ammonium formate buffer on all columns for the majority of solutes compared to formic acid (0.1% v/v). Ionogenic solutes were strongly affected by the choice of buffer, whereas neutral solutes were relatively unaffected (Fig. 3.2). The overall peak shape recovered when phosphoric acid (0.1% v/v) was used as mobile phase buffer (Fig.3.2a-c). There were also shifts in retention between the buffers, for example on the Cogent column bases were more strongly retained in formic acid than in ammonium formate buffer (data not shown). These phenomena were elucidated in detail in a collaborative study, discussed in 3.3. Page 52 of 246

54 Neutrals N (0.5) N (0.5) N (0.5) Bases (+ve charge) N (0.5) Acids (-ve charge) N (0.5) N (0.5) N (0.5) Fig. 3.2a Initial results for eleven probe compounds, Atlantis column. Vertical scale is peak efficiency at half-height in plates per column; peak asymmetry at 10% height shown in purple boxes above efficiency bar plots. Stationary phase Atlantis (4.6 x 250mm, 5µm) mobile phase % ACN with respective buffer. Blue = Ammonium Formate 5mM; Red = Formic Acid 0.1% v/v; Green = Phosphoric Acid 0.1% v/v

55 Neutrals N (0.5) N (0.5) N (0.5) Bases (+ve charge) N (0.5) Acids (-ve charge) N (0.5) N (0.5) N (0.5) Fig. 3.2b Initial results for eleven probe compounds, BEH Amide column. Vertical scale is peak efficiency at half-height in plates per column; peak asymmetry at 10% height shown in purple boxes above efficiency bar plots. Stationary phase BEH Amide (4.6 x 150mm, 3.5µm) mobile phase % ACN with respective buffer. Blue = Ammonium Formate 5mM; Red = Formic Acid 0.1% v/v; Green = Phosphoric Acid 0.1% v/v Page 54 of 246

56 Neutrals N (0.5) N (0.5) N (0.5) Bases (+ve charge) N (0.5) Acids (-ve charge) N (0.5) N (0.5) N (0.5) Fig. 3.2c Initial results for eleven probe compounds, Cogent Type C silica column. Vertical scale is peak efficiency at half-height in plates per column; peak asymmetry at 10% height shown in purple boxes above efficiency bar plots. Stationary phase BEH Amide (4.6 x 150mm, 3.5µm) mobile phase % ACN with respective buffer. Blue = Ammonium Formate 5mM; Red = Formic Acid 0.1% v/v; Green = Phosphoric Acid 0.1% v/v Page 55 of 246

57 Neutrals N (0.5) N (0.5) N (0.5) Bases (+ve charge) N (0.5) Acids (-ve charge) N (0.5) N (0.5) N (0.5) Fig. 3.2d Initial results for eleven probe compounds, ZIC-HILIC column. Vertical scale is peak efficiency at half-height in plates per column; peak asymmetry at 10% height shown in purple boxes above efficiency bar plots. Stationary phase BEH Amide (4.6 x 150mm, 3.5µm) mobile phase % ACN with respective buffer. Blue = Ammonium Formate 5mM; Red = Formic Acid 0.1% v/v. Page 56 of 246

58 3.3. Detailed studies to elucidate phenomena responsible for results in Comparison of performance of four different stationary phases with three different buffers The silica and bridged ethyl hybrid (BEH) amide phases were chosen for their, respectively, high and low cationic selectivity (preferential retention of cationic compounds) indicated in previous work (Kawachi et al. 2011, Kumar et al. 2013). The silica hydride stationary phase was studied as it supposedly contains very few silanol groups. Thus, ionic retention should be considerably reduced and its influence on retention and peak shape using FA would be considerably lower. A popular zwitterionic phase which has been used to separate hydrophilic species such as metabolites (Zhang et al. 2015) was also included as it can give alternative selectivity compared with the other phases in this work. The columns were evaluated using 90% ACN containing each of the three buffers AF, FA and PA, giving the results shown in Fig % water (90% ACN) was chosen for the study as it gave reasonable retention for most compounds (see also Section 3.3.2). The bare silica phase was characterised by low retention of the neutral solutes (uridine uracil) in both AF and FA (Fig. 3.3a). Retention of neutrals on this column was somewhat lower still in FA than AF, which might be attributed to a reduced water layer in the absence of the salt (Dinh et al. 2013). Greater retention of neutrals was obtained on both the zwitterionic and amide phases. This increased retention is likely to be related to the greater occupancy of the column pores with water on the ZIC phase (25%) and a TSK amide phase (21%) compared with Atlantis silica (9%) that have been measured using Karl-Fischer titration (data obtained in 80% ACN with 5 mm acetate, (Dinh et al. 2013)). This greater uptake of water is due to

59 the formation of swollen hydrogels on polymerically-functionalised phases like ZIC-HILIC (Dinh et al. 2011, 2013), and seems likely to emphasise the contribution of the partition mechanism to retention. With each mobile phase and each column, the retention of cationic solutes (nortriptyline to pyridine) was considerably higher than for neutrals. Preferential retention of cations compared with neutrals in AF or FA was shown for all columns, but was less pronounced on the amide phase. While the hydrophilicity of cations contributes to this increased retention, ionic interactions are likely to give a strong influence on retention. Thus nortriptyline (log D w w ph 3.0 = 1.1) is only moderately hydrophilic, but its high pka (10.2) results in protonation in all mobile phases leading to additional ionic retention. TMPAC (log D w w ph 3.0 = 2.2) is considerably more hydrophilic, which combined with similar ionic interactions (both nortriptyline and TMPAC are unipositively charged under the analysis conditions) leads to stronger retention than for nortriptyline. The retention of the weak base pyridine (pka 5.1, log D w w ph 3.0 = 1.0) was much greater in PA compared with AF. The lower ph of the PA mobile phase could result in greater protonation of this weak base in PA increasing its hydrophilicity and also increasing ionic retention caused by residual silanol ionisation (Fig. 3.3a). The persistence of ionic interactions on all columns at the low ph of PA is indicated by the observation that neutral uridine has a more negative log D w w ph 3.0 value ( 2.1) but considerably smaller retention in the PA mobile phase than pyridine ( 1.0). Besides the effect of the different ph values of these mobile phases on retention, the competing effect of the ammonium ionic retention should also be considered. This factor most likely contributes to the smaller retention of cationic solutes using AF compared with FA, which is particularly evident for the hydride, silica and zwitterionic columns (Fig. 3.3a). While the silica column is known to give high ionic retention (Dinh et al. 2011), the same result is surprising for the hydride column, which supposedly has hydride Page 58 of 246

60 groups in place of silanols. The same observation of high retention of bases on hydride phases, and the possibility of ionic retention of these solutes, has also been noted by other authors (Bawazeer et al. 2012). Similarly, the low retention of the anionic solutes such as 3,4,5-trihydroxybenzoicacid (pka 4.1, log D w w ph 3.0 = 0.60) and particularly of benzenesulfonic acid, BSA (pka 0.8, log D w w ph 3.0 = 2.0) on the silica and hydride columns (e.g. with AF) can be explained by repulsion of these species from ionised silanols. It seems that the positively charged ammonium ions cannot mask sufficiently the effects of the silanols, at least not with the low concentrations of AF used in this study. This repulsion must effectively counteract retention resulting from the partition mechanism, as BSA is appreciably hydrophilic. These anionic solutes only showed appreciable retention on the zwitterionic and amide phases, which have been shown to exhibit reduced ionic effects (Kumar et al. 2013). It is also possible that retention of acidic solutes is promoted by the presence of the quaternary ammonium functionality that is present on the zwitterionic phase. The continued low retention of BSA in PA on all columns is also suggestive of persistent silanol ionisation, even at lower ph. Fig. 3.3b indicates that the choice of buffer had relatively little effect on the efficiency of neutral compounds when 90% ACN was used (although lower efficiency was noted for some neutrals in FA using a lower water concentration 95% ACN see following section). The silica column gave excellent efficiency for almost all solutes (neutral, cationic and anionic) in AF buffer, generating >25,000 plates for some compounds (reduced plate height h < 2.0).High efficiency was also obtained for the amide column for all these solutes in AF, with h as low as 2.1 for the neutrals (note the dimensions and particle size of this column differed from the others). The lower plate count given by the zwitterionic column for neutrals was due to some peak tailing (see Fig. 3c), which influences the efficiency calculation, particularly when using the moments method. Page 59 of 246

61 While there was some decline in efficiency for the ionogenic solutes on all columns compared with the neutrals (especially for some of the cationic solutes with the hydride column, and for 3,4,5-THBA), efficiency was still broadly maintained at high levels for all columns when using AF. Clearly however, the most remarkable observation from Fig. 3.3b is the catastrophic loss in efficiency for the cationic solutes on all columns using FA, with plate counts in some cases only a tenth or less of their values in AF. Substantial deterioration in the efficiency of the acid BSA was also noted in FA; the high efficiency of benzoic acid in FA is attributable merely to the very low retention of this solute under the analysis conditions (see Fig. 3.3a). Clearly the problem with use of FA, noted previously only for a bare silica phase, is not connected merely with the strong ionic interactions of this material, as the amide phase (based instead on a hybrid organic inorganic silica) has considerably reduced interactions of this type (Kumar et al. 2013). Fig. 3.3c indicates that the loss in efficiency is in most cases due to serious fronting of peaks. However in ammonium formate buffers, the ammonium cation is presumably able to act as a counter-ion in cation exchange with basic compounds and additionally shield surface negative charges from acidic solutes. This perhaps explains the improved overall peak shape in AF as opposed to FA mobile phase (Fig. 3.3a). Page 60 of 246

62 Fig. 3.3a. Retention factor (k), for neutrals (uridine uracil), cationic(nortriptyline pyridine) and anionic solutes (benzoic acid-bsa) on four different columns using mobile phases with 90% ACN and various buffers. Solutes TMPAC = trimethylphenylammonium chloride; THBA = trihydroxybenzoic acid; BSA = benzenesulfonic acid. Column temperature 30 C. Solute concentration 50 mg/l, injection volume 5 µl. For other details, see Section 2. Page 61 of 246

63 Fig. 3.3b. Column efficiency (N, statistical moments method) for neutrals (uridine uracil), cationic(nortriptyline pyridine) and anionic solutes (benzoic acid-bsa) on four different columns using mobile phases with 90% ACN and various buffers. Solutes and conditions as per Fig 3.2a. Page 62 of 246

64 Fig. 3.3c asymmetry factor (As0.1) for neutrals (uridine uracil), cationic(nortriptyline pyridine) and anionic solutes (benzoic acid-bsa) on four different columns using mobile phases with 90% ACN and various buffers. Solutes and conditions as per Fig. 3.2a. Fig. 3.3 (a) Retention factor (k), (b) column efficiency (N, statistical moments method) and (c) asymmetry factor (As0.1) for neutrals (uridine uracil), cationic(nortriptyline pyridine) and anionic solutes (benzoic acid-bsa) on four different columns using mobile phases with 90% ACN and various buffers. Solutes TMPAC = trimethylphenylammonium chloride; THBA = trihydroxybenzoic acid; BSA = benzenesulfonic acid. Column temperature 30 C. Solute concentration 50 mg/l, injection volume 5 µl. For other details, see Section 2. Page 63 of 246

65 Fig ZIC-HILIC column (a) nortriptyline with mobile phase 90% ACN, 5 mm overall AF ph 3; (b) nortriptyline with 90% ACN containing 0.1% FA; (c) BSA with AF; (d) BSA with FA; (e) pyridine with AF; (f) pyridine with FA. Flow rate 0.5 cm 3 /min. Fig. 3.4 compares examples of the chromatograms for nortriptyline, BSA and pyridine in AF or FA mobile phase, showing peak fronting in the latter. Efficiency for cationic and anionic solutes was improved in PA compared with FA, but still inferior to that in AF for all columns. For this acid, either tailing or fronting caused loss inefficiency. Apparently, (partial) suppression of ionic interactions at low ph is not necessarily beneficial to obtaining good peak shapes for ionogenic solutes. Indeed it may be that the balance of ionic and Page 64 of 246

66 hydrophilic retention is the critical factor in determining peak shape. This balance may even be more favourable at higher ph (Periat et al. 2013c). While superior performance with AF may in part be related to the deactivating effect of the ammonium ion, it may be that the salt encourages the formation of the water layer on the column surface, giving improved results. The ionic strength of 0.1% FA in water is the least of the three in Table 3.1, but is likely to be considerably reduced in an 85% acetonitrile solution, as indicated by the rise in w s ph. The true thermodynamic s s ph (which pertains to the ph in the aqueous organic phase using calibration buffers prepared in the same solution) is related to the w s ph by the relationship (3.2): s s ph = s w ph δ (3.2) Where δ is a term that incorporates both the Gibbs free energy for transference of 1 mol of protons from the standard state in water to the standard state in the hydroorganic solvent at a given temperature, and the residual liquid junction potential (the difference between the liquid junction potential established during calibration in aqueous solutions, and that in the hydroorganic mixture). The value of δ is about 1.1 in 85% ACN (Gagliardi et al. 2007), implying s s ph = 4.0 and a concentration of formate anions in the mobile phase of only around 0.1 mm/l. While the w s ph of PA is lower than that of FA, the concentration of phosphate anions in the same mobile phase is still only around 0.8 mm/l. Still lower ionic strength would be present in solutions containing higher concentrations of ACN; the δ value in 90% ACN is about 1.6 (Gagliardi et al. 2007). Due (at least) to their low ionic strength, we believe that the degree of ion pairing in these simple acid solutions is likely to be small. In contrast, the ionic strength of the ammonium formate solution is maintained by the presence of the salt. In this case, some degree of ion pairing is possible, no studies are Page 65 of 246

67 known that have investigated this possibility in the high concentrations of ACN relevant to HILIC studies. It is possible that some ion-pairing in AF moderates ionic interactions with the stationary phase. Ion pairing would also reduce solute hydrophilicity: both effects may contribute to the reduced retention of bases shown in Fig. 3.3a. As mentioned previously however, all these arguments are complicated by the problem of whether physical parameters in water or in the aqueous organic mixture should be considered. The ionic strength of the mobile phase may well have influence on the thickness of the water layer which may be beneficial for ionic species. The presence of negatively-charged silanol groups on the stationary phase surface attracts cations, such the buffer cation [NH4] +. The cation itself is anticipated to be hydrated by some water molecules. Therefore when retaining on the stationary phase in HILIC, it is also possible that AF has specifically favourable properties in the formation of the water layer, in addition to its effect on masking ionised silanols. The observation of apparently strong ionic interactions of the hydride phase was unexpected compared with reports concerning the composition of this material (Boysen et al. 2011, Pesek et al. 2008). In fact in the present study, the hydride column appeared to behave in a fashion more similar to the bare silica phase, rather than phases like the zwitterionic and amide materials, which demonstrate reduced ionic interactions. Page 66 of 246

68 Fig. 3.5 Comparison of retention (k vs. k) plot for bare silica (Atlantis) vs. hydride silica (Cogent) using 90% ACN containing 5 mm ammonium formate w w ph 3.0. Other conditions as Fig Fig. 3.5 shows a correlation plot of k on the bare silica phase vs. the hydride silica phase, showing a high degree of correlation (r = 0.996), giving further evidence for their similar properties. In a previous publication (Kumar et al. 2013), the average correlation coefficient (r) of the retention factors of pairs of six different HILIC columns, again using a set of neutral, cationic and anionic solute sand similar mobile phase conditions to the present study, was This result emphasises the relative similarity of the hydride and silica phases, compared with the greater differences that typically exist between the selectivity of different HILIC stationary phases. Page 67 of 246

69 Effect of mobile phase water concentration and buffer on retention and peak shape The influence of 5 15% v/v water (95 85% v/v ACN) in mobile phases containing either 0.1% FA or 5 mm AF was studied on the bare silica and the BEH amide stationary phases, in order to investigate any possible variation in findings from the previous section when different water concentrations were used. The bare silica and amide phase were selected for this further study as they are indicated above and in (Kumar et al. 2013) to give strong and weak ionic retention, respectively. Page 68 of 246

70 b Fig. 3.6 a Fig. 3.6b Fig. 3.6a,b. Retention factor (k), column efficiency (N) and asymmetry factor (As0.1) measurements for Atlantis silica column using 85 95% ACN containing (a) 5 mm ammonium formate ww ph 3.0 (b) 0.1% formic acidasymmetry data in FA not shown for procainamide as split peaks were obtained. Other conditions see Fig Page 69 of 246

71 Fig. 3.6c Fig. 3.6d Fig. 3.6c,d. Retention factor (k), column efficiency (N) and asymmetry factor (As0.1) measurements for BEH amide column using 85 95% ACN containing (c) 5 mm ammonium formate w w ph 3.0 and (d) 0.1% formic acid; asymmetry data in FA not shown for procainamide as split peaks were obtained. Other conditions see Fig Page 70 of 246

72 Fig. 3.6 shows in all cases that retention increases substantially as the ACN content in the mobile phase is increased, which is in accord with increased partition into the stationary phase and/or increased adsorption onto polar surface groups, dependent on the separation mechanism. Considering the results in AF for the silica and amide columns, respectively (Fig. 3.6a and c), the increases in retention with increasing ACN concentration were particularly marked for the cationic solutes when using the silica column, suggesting a possible synergistic effect between hydrophilic and ionic retention. Increases in retention for these solutes using the amide column under the same conditions were considerably smaller. Fig. 3.6a shows that the high efficiency for most solutes in AF found in 3.2 when using 90% ACN was largely maintained over the range 85 95% ACN for the silica column. Average efficiency for the range of solutes in the buffer was again around 25,000 plates (h = 2.0). Poorer results were indicated again for 3,4,5-THBA, especially as the concentration of water decreased. The rapid decline of efficiency and increased asymmetry of 3,4,5-THBA in 95% ACN is attributable to the increased retention of this solute, which is very small at lower concentrations of ACN. It is possible that the increased contribution of strong adsorption of solute hydroxyl groups at high ACN concentration contributes to the deterioration in performance. Using the amide column with AF (Fig. 3.6c), decreases in efficiency at the lowest water concentration (95% ACN) were shown for the cationic solutes; 3,4,5-THBA was not eluted under these conditions. The decline in efficiency was accompanied by increased tailing of these solutes in 95% ACN. Using FA, little effect was observed on the efficiency of the neutral solutes on the silica column. However, a decline in efficiency at 95% ACN accompanied by increased tailing, was shown for these neutrals on the amide column (Fig. 3.6d). This decline may be connected with the decrease in the thickness of the water layer on the stationary phase as the concentration of water in the Page 71 of 246

73 mobile phase was reduced (McCalley et al. 2008b). Remarkable again is the drastic collapse in efficiency on both columns for the anionic and cationic solutes in FA compared with AF, which occurs over the whole range of water concentrations (Fig. 3.6b and d). Again, this drop in efficiency is caused mostly by serious fronting of the peaks in FA. This result once more indicates the necessity for use of buffers in order to achieve acceptable efficiencies for these solutes, even using a stationary phase like BEH amide that possesses reduced ionic interaction characteristics Causes of poor peak shape for cationic solutes in formic acid Fig. 3.7 shows a plot of column efficiency (using the statistical moments method) against sample load over the range 0.05 to 2.5 µg on column for the neutral uridine, the bases adenine and procainamide, and the quaternary compound TMPAC using the silica column and 90% ACN containing 0.1% FA as the mobile phase. It is clear that peak shape for uridine remains approximately constant, with the number of plates deteriorating by only around 13% over this range of sample load. In comparison, the deterioration in efficiency for adenine, procainamide and TMPAC was 61%, 93% and 96%, respectively. The drop in efficiency for these solutes was caused by increased fronting of the peaks as the sample load increased. It is possible that sparse cationic retention sites (ionised silanols) are increasingly overloaded by protonated solutes causing the deterioration in peak shape. As pointed out above, the ionic strength of 0.1% FA in 90% ACN is extremely low, and the concentration of mobile phase counterions (hydroxonium ions) may be insufficient to prevent solute interactions with these column groups. This situation would not arise in AF buffers, as the ionic strength is maintained by the presence of the salt. Page 72 of 246

74 Fig Effect of sample mass on efficiency of Atlantis silica column using procainamide (strong base), TMPAC (quaternary ammonium salt), adenine (weak base) uridine (neutral). Mobile phase 90% ACN containing 0.1% FA. Other conditions see Section 2. Page 73 of 246

75 Effect of buffer salt concentration and salt cation on retentionof cationic compounds Fig Effect of buffer salt on retention using Atlantis silica column. Mobile phase90% ACN containing salt adjusted to w w ph 3.0 with FA. Fig. 3.7 indicates that increasing the concentration of ammonium formate ph 3.0 from 5 10 mm in 90% ACN decreases retention for the strong bases nortriptyline and procainamide, and the quaternary compound TMPAC, showing the presence of ionic retention for these solutes on the silica column. Maintaining the buffer cation concentration at 10 mm by the substitution of rubidium cations for 5 mm of the AF concentration gave further decreases in the retention of these cationic solutes. It is well known that ion exchangers favour the bonding of ions of higher charge, decreased hydrated radius and Page 74 of 246

76 increased polarisability. Thus for the monovalent cations elution strength is generally in the order (Harris 2007): Cs+ > Rb+ > K+ > NH4+ > Na+ > H+ > Li+ Ions which are smaller in their non-hydrated state such as Li + have a higher charge density, attracting a larger number of water molecules, resulting in a larger hydrated radius. It is interesting that the weak base pyridine and also adenine show no evidence of ionic retention in Fig. 3.7, suggesting they are not protonated in the mobile phase. Pyridine has a pka of 5.1 in water, indicating that if ph values in water were applicable, it should be completely protonated in a mobile phase of w w ph 3.0. The w s ph of a similar mobile phase containing 85% ACN (Table 3.1) is 5.1, and combined with the effect of the depression of pka of bases in solutions of high organic solvent composition, would indicate that pyridine is not protonated, explaining the apparent lack of ionic retention and hydrophilic retention of this compound. The increased retention of these weak bases in 90% ACN containing 0.1% FA and PA as shown in Fig. 3.2a and discussed in Section 3.2 above, could be due to protonation of the compounds at the lower w s ph of these mobile phases compared with that of AF. These results indicate that w s ph should be considered in explaining results in HILIC despite the supposition of a layer of water on the surface of the phase. It is also possible that ionic interactions could occur between solutes situated in the bulk mobile phase and the column, in which case w s ph values would also be appropriate. Conclusions The retention and peak shape of some neutral, cationic and anionic solutes was investigated in aqueous acetonitrile mobile phases containing ammonium formate (AF), formic acid (FA) Page 75 of 246

77 and phosphoric acid (PA) on four HILIC columns of substantially different properties. Relatively little difference was found between these three mobile phases in terms of retention or column efficiency for the neutral solutes. While peak shapes of ionogenic solutes, particularly cationic compounds, were in general very good using AF, considerable deterioration in peak shape was observed when FA was used. The same result was obtained both on stationary phases with strong ionic retention characteristics (bare silica and hydride silica, which surprisingly showed very similar retention selectivity)and those exhibiting much lower ionic effects (hybrid silica amide and zwitterionic). Peak shape in FA became still worse as the sample load increased. Peak shape is likely to be related to the different ph and ionic strength of the various buffers, as measured in the aqueous or aqueous organic portion of the mobile phase. For example, the ionic strength of FA solutions in high concentrations of ACN is very low, and thus may adversely affect the formation of the water layer. In contrast, the presence of a reasonable concentration of ammonium ions is likely to encourage formation of the layer as well as masking some of the effects of ionic interactions. Ionic retention of bases was demonstrated by increasing the salt concentration, and by substitution of some of the ammonium for a rubidium salt, which in both cases reduced retention. (Partial) suppression of cationic retention on ionised silanol groups afforded by the use of low ph PA did not improve column efficiency compared with use of AF. Differences in the ph of the various buffers will affect the relative contribution of hydrophilic and ionic mechanisms to retention, which in turn may have an important influence on peak shape. Despite the supposition of a water layer on the column surface, the consideration of w s ph values seems important in explaining the retention of weak bases when using mobile phases rich in acetonitrile that are typical for HILIC separations. Page 76 of 246

78 Chapter 4 Performance of charged aerosol detection with hydrophilic interaction chromatography Page 77 of 246

79 Abstract The performance of the charged aerosol detector (CAD) was investigated using a diverse set of 29 solutes, including acids, bases and neutrals, over a range of mobile phase compositions, particularly with regard to its suitability for use in hydrophilic interaction chromatography (HILIC). Flow injection analysis was employed as a rapid method to study detector performance. CAD response was quasi-universal, strong signals were observed for compounds that have low volatility at typical operating (room) temperature. For relatively involatile solutes, response was reasonably independent of solute chemistry, giving variation of 12 18% RSD from buffered 95% ACN (HILIC) to 10% ACN (RP). Somewhat higher response was obtained for basic compared with neutral solutes. For cationic basic solutes, use of anionic reagents of increasing size in the mobile phase (formic, trifluoroacetic and heptafluorobutyric acid) produced somewhat increased detector response, suggesting that salt formation with these reagents is contributory. However, the increase was not stoichiometric, pointing to a complex mechanism. In general, CAD response increased as the concentration of acetonitrile in the mobile phase was increased from highly aqueous (10% ACN) to values typical in the HILIC range (80 95% ACN), with signal to noise ratios about four times higher than those for the RP range. The response of the CAD is non-linear. Equations describing aerosol formation cannot entirely explain the shape of the plots. Limits of detection (determined with a column for solutes of low k) under HILIC conditions were of the order of 1 3 ng on column, which com-pares favourably with other universal detectors. CAD response to inorganic anions allows observation of the independent movement through the column of the cationic and anionic constituents of basic drugs, which appear to be accompanied by mobile phase counterions, even at quite high solute concentrations. Page 78 of 246

80 1. Introduction An important problem for high performance liquid chromatography (HPLC) is the limited choice of detectors that respond to compounds containing no UV/VIS chromophores. Charged aerosol detection (CAD) is a relatively new type of detector developed for use in HPLC over the last 10 years (Dixon et al. 2002). About 100 publications concerning the detector have appeared to date (e.g. (Cohen et al. 2012, Gamache et al. 2005, Web of Science search topic Chromatograph* AND TITLE Charged Aerosol* )).The detector seems very suitable for the analysis of some pharmaceuticals and compounds of biomedical significance, at least in the reversed-phase (RP) mode (Vervoort et al. 2008), however, more detailed study is necessary to further understand its properties. Its response is dependent on the formation of aerosol particles (see Fig. 4.1), similar to techniques such as evaporative light scattering detection (ELSD) (Mourey et al. 1984) and condensation nucleation light scattering detection (CNLSD) (Allen et al. 1993). This dependence results in a response which is supposedly independent of solute molecular structure, giving a signal for any compound that is able to form stable aerosol particles. Therefore, CAD is potentially suitable for impurity analysis, particularly in pharmaceutical development where measurement by UV or mass spectrometry (MS) requires the use of standards that maybe unavailable for unknown impurities. In CAD, the aerosol particle becomes charged through collision with positively charged nitrogen gas (Vehovec et al. 2010), which differs from MS interfaces which generate molecular ions rather than charged particles (Niessen 2003). The present work aims to study the performance of the CAD, and investigate to what extent it may fulfil the requirements of a universal detector, particularly with regard to its use in hydrophilic interaction chromatography (HILIC). Clearly some factors influencing CAD behaviour are Page 79 of 246

81 already understood, although commercial instruments have some differences from the prototype described by Dixon and Peterson (Dixon et al. 2002). These differences are sometimes ignored in the literature in discussions of the mechanism of operation of commercial instruments (Almeling et al. 2012, Shaodong et al. 2010). Nevertheless, the process in both may involve transfer of charge from the sheath gas (e.g. nitrogen)to the solute particles (see Fig. 4.1), which is distinct from the more direct exposure of the corona discharge to the eluent as occurs in atmospheric pressure chemical ionisation (APCI) sources used in mass spectrometry. As CAD response (along with that of all aerosol detectors) depends on the formation of solid particles, it is limited to solutes that have low volatility at the operating temperature. However, few studies have investigated in detail any relationship between volatility and detector signal. The ability to differentiate between solute and mobile phase determines the detection limit, which has been quoted as ng sample on-column (Hutchinson et al. 2012, Cohen et al. 2012). Buffers are often critical additives to HPLC mobile phases in any separation mode, but are potentially detrimental to CAD performance. In HILIC, buffers can lead to better peak shape than simple acid solutions (McCalley 2007, Heaton et al. 2014c, Pesek et al. 2013), thus we wished to investigate their influence on CAD sensitivity. Furthermore, as with other aerosol-based detectors, detector response is dependent on organic solvent content. While changing detector response with organic solvent concentration has been investigated for its detrimental effect on response uniformity in gradient elution (Khandagale et al. 2014, Gorecki et al. 2006, Hutchinson, Li et al. 2010), high organic concentrations as used in HILIC may be advantageous for sensitivity as it should facilitate desolvation of particles in the CAD. Aerosol-based detectors are known to produce non-linear calibration curves (Hutchinson et al. 2011), which can arise for different reasons in different detectors. For instance in the ELSD, it is due to both the non- Page 80 of 246

82 linearity of aerosol formation and a change in detection mechanism with the size of aerosol particles (Stolywho et al. 1983). The mechanism of detection in CAD is more straightforward than ELSD (Vervoort et al. 2008), and CAD calibration curves can be close to linear over small concentration ranges (Vehovec et al. 2010). The detailed mechanism that causes non-linearity of CAD calibration curves and their profile has not been described to date. Detector response for aerosol-based detectors is believed to be mostly independent of solute chemistry (Vervoort et al. 2008). However this factor has also not been investigated in much detail with respect to CAD for a sufficiently broad selection of solute structures. Approximately 50% of drug active pharmaceutical ingredients (API) are salts (Paulekuhn et al. 2007), and many salt counter ions do not contain chromophores. An important benefit of CAD is the ability to detect solutes which do not contain chromophores, and thus it should respond to these counterions (Schiesel et al. 2012). Page 81 of 246

83 Fig. 4.1 Simple Schematic of CAD operation 2. Experimental 2.1 Chemicals and reagents A set of 29 probe compounds comprising acids, bases and neutrals (as used in a previous study (Kumar et al. 2013)) was obtained from SigmaAldrich (Poole, UK) and used as probes. Structural and physico-chemical data are provided in Table 4.1. Log D values were calculated as the average from three different software packages: ACD version 12.0 (ACD Labs, Toronto, Canada), Marvin (Chem Axon, Budapest Hungary) and MedChem Designer (Simulations Plus, Lancaster, USA). Standards were diluted in the exact mobile phase from stock solutions typically at 10,000 mg/l made up in 50% ACN containing 0.1% FA. ACN (HPLC gradient grade), ammonium formate (AF), formic acid (FA) (LCMS grade), ammonium acetate (AA) and acetic acid (HPLC grade), were purchased from Fisher Scientific (Loughborough, UK). Page 82 of 246

84 Table 4.1. Identities, structures and physico-chemical characteristics of test compounds. Solute Structure MW FW Log D (ph 3)*** BP / C MP / C 4-hydroxybenzoic acid * Caffeine Diphenhydramine * 168# 3,4,5 THBA * BSA ** 65.5 BTEAC Procainamide * 167# TMPAC Page 83 of 246

85 Solute Structure MW FW Log D (ph 3)*** BP / C MP / C Cytidine * 225 Phluroglucinol * ,4 dihydroxypyridine * NSA ** Nortriptyline * 214$ 2 -deoxyuridine ** 165 Theophylline * 272 2,3 dihydroxypyridine * 245 Pyridine Page 84 of 246

86 Solute Structure MW FW Log D (ph 3)*** BP / C MP / C Benzoic acid Cytosine ** Paracetamol * Thiourea * 177 Uridine ** 165 2,3 dihydroxybenzoic acid ,6 dimethylpyridine BTMAC N,N Dimethylacetamide Adenine * 220 Page 85 of 246

87 Solute Structure MW FW Log D (ph 3)*** BP / C MP / C Uracil ** 335 Theobromine ** 357 * predicted at 760 mmhg using ACD labs program (see Experimental). ** predicted from ( *** average log D from three packages (see Experimental) # value from ( $ value from ( 2.2 Equipment and methodology A Thermo UltiMate 3000 Rapid Separation Liquid Chromatography system was used for all experiments, comprising a quaternary pump, diode array detector (DAD) and either a Corona Ultra or Corona Veo CAD, with Chromeleon 7.2 software (Thermo, Germering, Germany). The CAD is a destructive detector, therefore the DAD and CAD detectors were connected in series in some experiments, with flow first through the DAD. Thermo Viper tubing (0.13 mm ID) was used as connection tubing. Data collection rates were 100 Hz for both DAD and CAD, due to narrow peak widths (typically 1 s at half height in flow injection analysis (FIA)). The Corona Ultra nebuliser (cross flow design similar to that used in atomic Page 86 of 246

88 absorption spectrometry) was controlled at 22 C with the evaporator tube at ambient temperature, while the Veo (concentric flow design similar to those used in mass spectrometry) nebuliser was at ambient temperature and the evaporator tube set to 30 C. The Veo had a power function (PF) designed to linearise data, which was set to either 0.67 (this simulates off ), 1.00 (the default) or 1.2 (optimised setting using experimental data, see below). An ethylene bridged hybrid (BEH) amide column ( mm, particle size = 3.5 µm, Waters, Milford, USA) was used for determination of the detection limit, linearity and for the salt separation experiments. An Atlantis bare silica column ( mm ID, particle size = 5 µm, Waters) was used for some salt composition experiments. The mobile phase was ACN-5 mm ammonium for-mate or ammonium acetate buffer (80:20, w/w) unless otherwise stated. The ph meter was calibrated in aqueous buffers and formic or acetic acid was used to adjust the aqueous portion to w w ph 3 or 5.Solutions at w w ph 6.8 were unadjusted 5 mm ammonium acetate. Care is necessary as ph calibration buffers can be a major source of non-volatile contaminants in the mobile phase. In flow injection analysis (FIA), narrow bore tubing (75 µm 1100 mm) was used in place of the chromatographic column to maintain sufficient backpressure. Samples for FIA were prepared at a concentration of 300 mg/l; injection volumes were1 µl unless otherwise stated. Flow rate was 1 ml/min. For calculation of retention factors, toluene is generally used as a void volume marker in HILIC with UV detection (Heaton et al. 2014b), but is too volatile for use with the CAD. Naphtho [2,3-a] pyrene appeared to be a suitable alternative for CAD. Page 87 of 246

89 3. Results and discussion Detection limits (HPLC) When applied to the impurity profiling of amino acid mixtures in nutritional infusion bags, CAD limits of quantitation (LOQ) were reported at 10 ng on-column (1 µg/ml; 10 µl injection) (Schiesel et al. 2012). The authors used a signal to noise ratio of 5:1 to determine the LOQ (Schiesel et al. 2012), whereas common practice is to use a S:N of 10. Ramos et al. reported CAD limits of detection (LOD) 4 times lower than ELSD for analysis of membrane phospholipids by normal phase HPLC (Ramos et al. 2008). Hutchinson et al. reported that the LOD for 11 solutes was over 5 times smaller using CAD compared with ELSD (Hutchinson et al. 2011). Detection limits for an acid, neutral and basic solute in our experiments are shown in Table 4.2 for a typical HILIC mobile phase (5 mm ammonium formate ph 3 in 80% ACN). A signal to noise ratio of 3 was used as LOD and 10 for LOQ. The LOD of 1 3 ng and LOQ of 5 9 ng (both on column, 1 µl injections) compare favourably with other universal detectors such as refractive index (LOD 1 µg on column (Yeung et al. 1986)) and ELSD (LOD ng on column (Vervoort et al. 2008, Shaodong et al. 2010)). While the data in Table 4.2 was recorded for the BEH column, using the same mobile phase we did not observe serious noise or bleeding with the Atlantis column, as reported by Jia et al. (Jia et al. 2011). Table 4.2 (and indeed most of this work) was based on use of an acidic mobile phase, whereas Jia et al. used unbuffered ammonium acetate that has approximately neutral ph in aqueous solution. This higher ph might have caused some dissolution of silica and thus noise in the CAD. Page 88 of 246

90 Table 4.2. Detection limits for charged aerosol detection in HILIC conditions. HPLC, mobile phase 80% ACN, 5mM ammonium formate w w ph 3. Solute LOD / mg per L LOQ / mg per L BSA 3 9 Uridine 2 6 Nortriptyline 1 5 Page 89 of 246

91 3.1.2 Calibration curves (HPLC) It has been reported that over wide ranges of analyte concentration, (e.g ppm) CAD response is non-linear, while over narrow ranges of analyte concentration, it is quasi-linear (Vehovec et al. 2010). Using FIA, Hutchinson et al. investigated CAD calibration using sucralose, amitriptyline, dibucaine and quinine at concentrations of 1 µg/ml to 1 mg/ml (25 µl injections) (Hutchinson et al. 2010). Although not commented on by the authors, their data suggest a low quasi-linear range below approximately 0.05 mg/ml and an upper quasilinear range between 0.4 and 1.0 mg/ml. Fig. 4.2 shows calibration plots for the acid BSA, the base nortriptyline and the neutral uridine over the range mg/l, using a BEH amide column with the CAD (Fig. 4.2a) and for UV detection (Fig. 4.2b). Hutchinson et al. (Hutchinson et al. 2010) reported maximum reliable CAD response at 70% ACN, and recommended this concentration for applications requiring maximum sensitivity. Retention is often poor at this ACN concentration in HILIC; a typical range for HILIC is 70 95% ACN. We therefore selected 80% ACN for our study. UV detection shows excellent linearity over the entire range (lowest R , for Nortriptyline). Our results indicate also a lower quasilinear range (1 100 mg/l) and an upper quasi-linear range ( mg/l) for CAD. The calibration curves appear to be sublinear (concave to the x-axis). The consistent general shape of CAD calibration curves (Fig. 4.2a; also (Hutchinson et al. 2010)) is perhaps described by some empirical formula, however no such interpretation has been attempted to date. Page 90 of 246

92 DAD peak area / mau*min Ultar peak area / pa*min BSA Uridine Nortriptyline concentration mg / L Fig. 4.2 a. Figure 4.2. Peak area vs. concentration for a neutral (Uridine), acid (BSA) and base (Nortriptyline) (a) CAD Ultra (HPLC, mobile phase 80%ACN, 5 mm ammonium formate ph 3). 30 R² = R² = R² = concentration mg/l Fig. 4.2 b Figure 4.2. Peak area vs. concentration for solutes as per Fig. 4.2a, (b) DAD. Conditions as per Fig. 4.2a. Page 91 of 246

93 Ultra Peak Area / pa*min log (Ultra peak area) R² = R² = R² = log (concentration) Fig. 4.2 c. Figure 4.2. Peak area vs. concentration for solutes as per Fig. 4.2a, (c) log/log CAD Ultra. Conditions as per Fig. 4.2a R² = Density / g cm -3 Fig. 4.2 d. Peak Area vs. Density for 17 non-volatile solutes (FIA, mobile phase 80% ACN with 5mM ammonium formate ph 3). Page 92 of 246

94 Figure 4.2. Peak area vs. concentration for a neutral (Uridine), acid (BSA) and base (Nortriptyline) (a) CAD Ultra, (b) DAD and (c) log/log CAD Ultra (HPLC, mobile phase 80%ACN, 5 mm ammonium formate ph 3); (d) Peak Area vs. Density for 17 non-volatile solutes (FIA, mobile phase 80% ACN with 5mM ammonium formate ph 3). The particle size dp as described for ELSD (1) is given by the equation (4.1): dp = dd ( C ρ p ) 1/3 (4.1) where dd is the diameter of the droplet, c is the analyte concentration and p is the density of the particle. The aerosol formation step should be similar in ELSD and CAD. It was assumed by Charlesworth that aerosol particles are approximately spherical for ELSD (Charlesworth 1978), as confirmed by fundamental studies (Reid et al. 2011). Dixon and Peterson assumed this for their prototype aerosol detector (Dixon et al. 2002). The authors of that study reported that the prototype s detector sensitivity (defined as the gradient of the calibration curve) was lower for solute particles with diameters greater than 10 nm, i.e. the gradient of a plot of response vs. concentration is steep at low concentrations and becomes shallower at higher concentrations. The surface area of a sphere (A) is given by (4.2): A = 4 πr 2 (4.2) It follows that: Ap = dd 2 π ( C ρp) 2/3 (4.3) Page 93 of 246

95 Thus, the surface area of the particle (Ap) is theoretically proportional to solute concentration via the power equation (4.3). Therefore an increase of solute concentration results in a larger particle surface area and higher detector response. The exponent of (4.3) is not unity. This theory assumes that adsorption of charged nitrogen onto the particles is a linear (Langmuirian) relationship with surface area. Equation (4.4) simplifies the relationship between CAD response and analyte concentration (c). Plotting CAD response vs. concentration should yield a non-linear curve with a fractional exponent of 2/3.This relationship is in agreement with work from the manufacturer of the charged aerosol detector (Thomas et al. 2014). CAD Response C 2/3 (4.4) Taking logs gives (4.5), a simple linear relationship, where log of the coefficient a becomes the intercept and the slope is 2/3. log(cad Response) = 2 log C + log a (4.5) 3 Fig. 4.2c shows log/log calibration plots for Nortriptyline, BSA and Uridine. Linearity was very good, much-improved compared to the raw data (Fig. 4.2a) giving R 2 values of ; other authors have noted similary good linearity of these log/log plots (Ramos et al. 2008, Eom et al. 2010). The gradient of these plots ranged from to 0.976, (as expected from the sublinear nature of Fig. 4.2a) which clearly is larger than the value of 2/3 expected from equation (4). Chaminade et al. reported gradients of CAD log/log calibration plots between 0.79 and 1.11 for membrane phospholipids using normal phase separation (Ramos et al. 2008). Nevertheless, log/log plots seem a pragmatic way to calibrate the detector. Newer CAD models such as the Corona Ultra RS and Corona Veo contain an in-built power function feature, which is intended to linearise data. This function appears to be based Page 94 of 246

96 broadly on the arguments presented above, although its operation is proprietary. The use of a user-inputted power function of 1.2 on the Corona Veo gave linearity similar to the log/log plots discussed. Density is also a factor affecting solute particle size in equation (4.3), therefore the possible effect of solute density on detector response was investigated using flow injection analysis data gathered later in this study (see 3.2.3). A plot of detector response vs. solute density is shown in Fig 4.3 d. No apparent relationship was indicated by the coefficient of determination (R 2 = ). However the range of densities in Fig. 4.3d is somewhat small ( gcm -3 ). It is speculated that the aerosol particles pack looser at high solute concentrations, leading to a low density which perhaps results in larger than expected particles and a response that is closer to linear than equation (4.3) suggests. To qualify this would perhaps require fundamental studies into measuring aerosol particle density at increasing solute concentration, which is outside the scope of this project. Page 95 of 246

97 3.2 Response universality and uniformity Flow injection analysis FIA is a rapid method of determining detector response, avoiding any problems of interference from the column (e.g. irreversible adsorption of part of the injected solute). Problems have been reported of poor reproducibility of peak area at low solute concentrations by FIA (Gorecki et al. 2006). It appears that this problem may be related to disturbances shown in blank injections of pure mobile phase. These were minimised by using analyte concentrations of 300 mg/l, which gave blank disturbances that were very small in comparison with the analyte signal Effect of solute salt composition on response (HPLC; FIA) To obtain net neutrality, ionised solutes must be associated with a counterion. This counterion is usually assumed to originate from the mobile phase, although it is conceivable that the counterion from the injected salt is involved, dependent on solute concentrations and mobile phase conditions. To investigate this further, the chloride, bromide and iodide salts of the quaternary ammonium compound benzyltriethylammonium (BTEA) were prepared at 300 mg/l of salt (e.g. 300 mg/l of BTEAC), and individually separated by HPLC using a BEH Amide column (80% ACN, 5 mm AF ph3) (Fig. 4.3a c). The salts were also analysed by FIA, with identical mobile phase. For analysis by HPLC, peak areas for the same injected mass of each salt decreased for the cationic moiety in the order BTEAC > BTEABr > BTEAI (Table 4.3). This result is in agreement with the cationic part of the salt contributing a decreasing fraction of the total mass as the anion gets larger (chloride to iodide). For HPLC Page 96 of 246

98 analysis, the result indicates that the solute cation maybe accompanied by formate anions during passage through the column, as the injected solute cation and anion clearly separate on the column (Fig. 4.3a c). Note also the different retention times of chloride, bromide and iodide in these Figures. However, the same pattern of detector response was found by FIA, which involves no separation process. The standards were first diluted from the stock solutions (10, mg/l) with the FIA/HPLC mobile phase and injected into the same solution. This suggested that the large excess of sample diluent formate anions (see Section 2) and mobile phase buffer anions (5 mm AF ph 3) largely replace the solute (halide anions) in solution, which may influence the subsequent formation of aerosol particles. Thus the solute halide ions may have little contribution to the overall response even in FIA. A loading study for nortriptyline hydrochloride was carried out to further investigate the separation of the anion and cation in the HPLC process, up to much higher concentrations than those used in FIA. The salt was dissolved in the exact mobile phase at concentrations from 100 to 10,000 mg/l; separations were carried out on the BEH Amide and also on an Atlantis silica column. Fig. 4.3d and e show distinct peaks for the nortriptylinium cation and the chloride anion at all concentrations on the amide and bare silica column respectively. In the separation of amino acids by electrostatic repulsion-hydrophilic interaction chromatography (ERLIC), Alpert (Alpert 2008) showed a symmetrical peak for arginine when the solute was dissolved in mobile phase 10 mm triethylaminephosphate (TEAP) ph 2 in 70% ACN, using a Polywax LP column. The peak was attributed to arginine phosphate. However, when the solute was dissolved instead in triethylaminemethylphosphonate (TEA-MePO3) an additional peak appeared at earlier retention time, with a continuum evident between the peaks. It was suggested that the earlier peak was due to arginine molecules that had retained MePO3as the counterion, while some slow counterion exchange takes place with Page 97 of 246

99 the mobile phase. Data in Fig. 4.3 show a different behaviour, as they suggest independent migration of the solute anion and cation through the column. Clearly, the result is likely to be influenced by the exact combination of solute, mobile phase buffer (and their concentrations), and stationary phase used. Fig. 4.3d shows detector overload for the modestly-retained nortriptyline (k = 1.9) above 20 µg sample load on the BEH amide column, but no evidence of detector overload for the well-retained chloride (k = 21) even at 100 µg sample load. The chloride peak continues to increase in size even at the highest sample loads. Fig. 4.3e, using the Atlantis column, shows detector overload for the nortriptylinium cation only above 80 µg sample load and none for the chloride anion, with peaks continuing to behave independently. These data suggest a detector dynamic range of 1 ng to over 20 µg, i.e. over 4 orders of magnitude. The results indicate that even at the highest concentrations studied, a plateau in the chloride response is not attained, suggesting that no association of chloride with nortriptyline occurs. The elution order of solute anion and cation reversed when switching from the BEH Amide (Fig. 4.3d) to the Atlantis bare silica column (Fig. 4.3e), due to much stronger cation exchange retention of nortriptyline on the Atlantis column (Kumar et al. 2013). Page 98 of 246

100 Fig. 4.3a. HILIC-CAD separation and detection of the salt (a) benzyltriethylammonium chloride. Peak identities 1 = benzyltriethylammonium, 2 = chloride. HPLC, BEH Amide column, mobile phase 80% ACN with 5 mm ammonium formate ph mg/l, injection volume 1 µl. Fig. 4.3b. HILIC-CAD separation and detection of the salt (b) benzyltriethylammonium bromide. Peak identities 1 = benzyltriethylammonium, 3 = bromide. HPLC, BEH Amide column, mobile phase 80% ACN with 5 mm ammonium formate ph mg/l, injection volume 1 µl. Page 99 of 246

101 Fig. 4.3c. HILIC-CAD separation and detection of the salt (b) benzyltriethylammonium iodide. Peak identities 1 = benzyltriethylammonium, 4 = iodide. HPLC, BEH Amide column, mobile phase 80% ACN with 5 mm ammonium formate ph mg/l, injection volume 1 µl. Fig. 4.3d. HILIC-CAD separation and detection of the salt (a) nortriptylinium chloride. Peak identities 2 = chloride, 5 = nortriptylinium. HPLC, Atlantis column, mobile phase 95% ACN with 5 mm ammonium formate ph 3. Nortriptyline hydrochloride concentration ,000 mg/l, injection volume 10 µl. Page 100 of 246

102 Fig. 4.3e. HILIC-CAD separation and detection of the salt (a) nortriptylinium chloride. Peak identities 2 = chloride, 5 = nortriptylinium. HPLC, Atlantis column, mobile phase 95% ACN with 5 mm ammonium formate ph 3. Nortriptyline hydrochloride concentration ,000 mg/l, injection volume 10 µl. Figure 4.3. HILIC-CAD separation and detection of the salts (a) benzyltriethylammonium chloride, (b) benzyltriethylammonium bromide, (c) benzyltriethylammonium iodide; (d) (e) nortriptyline hydrochloride. Peak identities 1 = benzyltriethylammonium, 2 = chloride, 3 = bromide, 4 = iodide, 5 = nortriptylinium. HPLC, mobile phase 80% ACN for(a) (c), 95% ACN for (d), 90% ACN for (e) all containing 5 mm ammonium formate ph 3, Atlantis column for (e), BEH Amide column for all others. Nortriptyline hydrochloride concentration ,000 mg/l, injection volume 10 µl, others 300 mg/l, injection volume 1 µl. Page 101 of 246

103 Table 4.3. Peak areas of BTEAC, BTEABr and BTEAI by FIA and HPLC. Mobile phase 80% ACN, 5mM ammonium formate w w ph 3. Compound Peak area of BTEA+ (FIA) Peak area of BTEA+ (HPLC) Proportion of base BTEAC % BTEABr % BTEAI % Page 102 of 246

104 3.2.3 Response Uniformity (FIA)-dependence on solute and mobile phase buffer CAD peak areas were measured for injection of 300 ng of the 29 compounds in 80% ACN, 5 mm AF ph 3 using FIA (Fig. 4a). The relative standard deviation (RSD) of the response for 21 compounds (omitting 8 with no or low response N,N-dimethylacetamide to caffeine) was 14% in this mobile phase, which shows reasonable uniformity considering the diverse structures of the compound set. Greater response uniformity for CAD in comparison with UV detection is also seen in Fig. 4.2a and b. Response appeared somewhat higher for basic compounds (shown in blue) than neutrals (green), albeit with some overlap (Fig. 4a). This observation is unexpected and unreported to date, as the production of physical particles by aerosol-based detectors should be independent of solute chemistry. Ionogenic compounds are often available in their salt form (e.g. Nortriptyline 300 mg/l was prepared as 300 mg/l of Nortriptyline HCl salt). While neutral compounds would not be expected to interact strongly with mobile phase buffer constituents in particle formation, this is clearly a possibility for ionogenic compounds, and may be responsible for the differences in response. The mean response for the same 21 compounds was compared for additives commonly used in HILIC in addition to ammonium formate (AF) including formic acid (0.100%, v/v) (FA), ammo-nium acetate (AA), trifluoroacetic acid (0.200%, v/v) (TFA), and heptafluorbutyric acid (0.345%, v/v) (HFBA) (Table 4.4). The acid solutions were equimolar (26.5 mm). The mean response at 80% ACN concentration did not appear to be greatly affected when changing the ph of the buffer, or simple acid modifiers. However, the spread of response to the individual compounds was greater in mobile phases of equimolar TFA and HFBA (23% and 30% RSD, respectively) than for the other mobile phases. For the particular case of ionised solutes, association with mobile phase counterions of increasing Page 103 of 246

105 mass might be expected to produce an increase in CAD response, giving some explanation of these data. Thus for analysis of cationic solutes, the response might increase in the order for-mate, trifluoroacetate (TFA) and heptaflurobutyrate (HFBA) whose anions have molar mass 45, 113 and 213, respectively. Koupparis and co-workers (Galanakis et al. 2006) found that the response for the hydrophilic antibiotic Amikacin using ELSD increased for TFA over FA, further increasing for higher concentrations of TFA. In addition, they evaluated HFBA and nonafluoropentanoic acid (NFPA), claiming that their ELSD responses increased in relation to the mass of the anion of these strong acids. However, the concentration of acid was not kept constant between datasets, either in terms of v/v or molar concentration. If response indeed increases in proportion to the added mass of a heavier anion, the ratio of response for strongly basic solutes,e.g. nortriptyline should be in the order nortriptyline formate: nortriptyline trifluoroacetate: nortriptylineheptafluorobutyrate 1.00:1.22:1.54. To investigate this hypothesis, FIA was performed in FA, TFA and HFBA each at a concentration of 26.5 mm. The compounds used were the stronger base nortriptyline, the weak base cytosine, neutral uridine as a control, the strong acid benzenesulfonic acid and the weak acid 4-HBA. A solute concentration of 600 mg/l was used, which is somewhat higher than used above because of higher baseline disturbances from TFA and HFBA compared to FA. The results (Fig. 4b) show that nortriptyline response did not follow the predicted ratios, with response of 1.00:1.03:1.13 for the FA, TFA and HFBA, respectively. The weak base cytosine followed the same trend, with response increasing in the ratio 1.00:1.09:1.33 for FA, TFA and HFBA but not in line with the predicted increase of 1:1.43:2.07.It is apparent from Fig. 4a that neutral solutes have broadly lower response than ionised basic solutes, although the cause of this is unexplained to date. There have been no published reports of mobile phase ph affecting small molecule solute CAD response, therefore the weak acid 4- Page 104 of 246

106 HBA was expected to also be unaffected by choice of acid buffer. However, Fig. 4b shows a decrease in response for 4-HBA in both TFA and HFBA compared with FA (33%decrease in TFA compared to FA). TFA and HFBA are capable of neutralising 4-HBA under the conditions used; decreasing the degree of ionisation of 4-HBA in the stronger acid would result in a predominantly neutral form of the solute reaching the detector. BSA, which is deprotonated at all ph values, was unaffected by the choice of acid buffer. Uridine, which is neutral under these conditions showed small reductions in response in TFA and HFBA (less than 10% reduction in peak area). Khandagale et al. described the CAD solute plug as a plume (Khandagale et al. 2013), which travels within the detector after nebulisation. The plume has only a finite period of time to undergo all the processes required to produce a peak in the CAD (Fig. 4.1). We estimated the detector residence time at 1 s using an effective detector volume of 14 µl reported by the manufacturer (Gamache et al. 2005) and a flow rate of 1 ml/min. The process of forming aerosol particles by evaporation of aerosol droplets is possibly analogous to crystal formation from bulk solution. Ionic solids have much higher melting points than solids of neutral compounds and it is well-known that optimum growth rates for ionic crystals are at least a factor of 10 times greater than for molecular crystals, due to the high strength of coulombic intermolecular interactions relative to weaker van-der-waals and London dispersion forces (Wright 1989). Perhaps ionogenic solutes are better able to form stable aerosol particles within this short time window, compared to neutrals which are held together by weaker interactions. Page 105 of 246

107 Fig. 4.4a. CAD response for 29 compounds (FIA, mobile phase 80% ACN, 5 mm ammonium formate ph 3). Blue = bases, red = acids, green = neutrals. 1 µl injections Page 106 of 246

108 Fig. 4.4b. CAD Ultra response in dilute acids for the bases nortriptyline and cytosine, acids BSA and 4-HBA and the neutral Uridine (FIA, 80% ACN, FA (0.1% v/v) vs. TFA (0.2%) vs. HFBA (0.345%)). Predicted values from ratios explained in in hashed-line bars. Figure 4.4. (a) CAD response for 29 compounds (FIA, mobile phase 80% ACN, 5 mm ammonium formate ph 3). Blue = bases, red = acids, green = neutrals. (b) CAD Ultra response in dilute acids for the bases nortriptyline and cytosine, acids BSA and 4-HBA and the neutral Uridine (FIA, 80% ACN, FA (0.1% v/v) vs. TFA (0.2%) vs. HFBA (0.345%)). Predicted values from ratios explained in in hashed-line bars. Page 107 of 246

109 3.2.4 Effect of solute volatility on response The CAD response using FIA for the set of 29 diverse compounds (Fig. 4a) indicates that eight gave low or no response, three of which were liquids at room temperature (pyridine, 2,6-dimethylpyridineand N,N-dimethylacetamide). Solutes which respond poorly in CAD are too volatile to form stable aerosol particles (Gamache et al. 2005). Some relationship between solute volatility and response might be expected. Compounds that respond to CAD in general are those which have higher boiling point, melting point or molecular mass (Fig. 5a c respectively), which are typical indicators of solute volatility. However, there are clear exceptions. 2,3-dihydroxybenzoic acid (bp 344 C, mp 205 C) gives only a third of the response of diphenhydramine (bp 344 C, mp 168 C). Benzoic acid has an appreciably high boiling point at 249 C and molecular mass of 122 g/mol but gave no response whatsoever; thiourea is smaller with even lower boiling point (bp 187 C, MW 76 g/mol, mp 177 C) but its CAD response was strong. However, benzoic acid has a low mp (122 C) and is known to be volatile, sufficiently so that its analysis by headspace Gas Chromatography (GC)-MS is possible (Pellati et al. 2013). The exceptions make a definitive cut-off point for a response difficult to predict purely from physico-chemical indicators of solute volatility. A minority group of four solutes including caffeine and 4-hydroxybenzoic acid (4-HBA) responded in CAD, but with peak areas ca. 40% lower than the strong CAD responders. The bp and mp points of these solutes are diverse. Indeed the data for 4-HBA (bp 336 C, mp214 C), 2,3- dihydroxypyridine (bp 441 C mp 245 C), and 2,3-dihydroxybenzoic acid (bp 344 C, mp 205 C) overlap with those for strong responders. Caffeine is frequently referred to as semivolatile (Lauritsen et al. 1997). It has a sublimation point (bp Table 4.1) of 178 C, considerably below its melting point of 238 C ( which may explain Page 108 of 246

110 its behaviour, although its sublimation temperature is still well above the detector settings. The CAD nebuliser is set to room temperature by default, and the evaporation process is also endothermic. Therefore this definition perhaps does not apply to CAD. It seems that experimental measurement (e.g. by FIA) is necessary to confirm response. Fig. 5a c show clearly that no relationships are apparent between response and melting point, molecular mass, and boiling point, with correlation coefficients close to zero. This result however fits with the claim of reasonably uniform response for non-volatile substances for the CAD. For ELSD, which might be expected to show similar effects, conflicting findings have been published on the effect of MW on detector response (Stolywho et al. 1983, Mourey et al. 1984). Page 109 of 246

111 Figure 4.5. CAD response for 29 compounds, plotted against (a) boiling point, and (b) melting point; (c) molecular mass (FIA, conditions as per Fig. 4.2). Page 110 of 246

112 3.2.5 Effect of organic modifier (FIA) Haddad and co-workers showed that CAD signal increases in proportion to the organic solvent concentration of the mobile phase for acetonitrile, acetone, isopropyl alcohol and methanol (Hutchinson et al. 2012, Hutchinson et al. 2010). These previous studies considered only peak areas and not the effect on uniformity of response or signal to noise ratio. Fig. 4.6a shows a plot of peak area vs. organic solvent concentration for 10 95% ACN. CAD response is roughly proportional to ACN con-centration, in good agreement with earlier reports (Hutchinson et al. 2012, Hutchinson et al. 2010). Our data show peak areas under typical HILIC conditions (70 95%ACN) roughly twice that of typical RPLC conditions (10 50% ACN).Excessively large droplets are removed by droplet selection inside the nebuliser (Fig. 4.1). It is possible that highly-aqueous mobile phases produce excessively large droplets by condensation, which are removed in the nebuliser, resulting in reduced CAD response. Smaller droplets in greater numbers are formed with the lower viscosity, density and surface tension of highly organic eluents (Khandagale et al. 2014), which perhaps explains the better transport efficiency in these conditions. Table 4.4 shows that with AF ph 3 as buffer the uniformity of response was slightly improved in HILIC conditions compared with RPLC. Moreau found CAD background current was higher for organic solvents compared to water (Moreau 2006), attributable to their dry residue content (typically 2 ppm); it is therefore conceivable that the high ACN content might also cause high noise with HILIC mobile phases. We measured noise over a 30 min period after system stabilisation (Fig. 4.6b); noise was lower in higher ACN concentrations. As a result, the response of the CAD (measured in terms of S/N) was further improved (Fig. 4.6c) under HILIC conditions compared with the improvements in terms of crude solute Page 111 of 246

113 peak area. It is possible that while organic solvents can indeed be a source of particulates, the fine aerosol particles are sufficiently small in organic-rich mobile phases (Khandagale, et al. 2014) that they are poorly detected. Table 4.4. Peak areas and uniformity of response for 21 compounds in a selection of HILIC mobile phases. Mean Ultra response / pa*min 95% ACN 5mM AF ph3 80% ACN 5mM AF ph3 50% ACN 5mM AF ph3 10% ACN 5mM AF ph3 Mobile phase 80% 80% ACN ACN 5mM 5mM AF AA ph 5 ph 5 80% ACN 5mM AA ph % ACN FA (0.1% v/v)* 80% ACN TFA (0.2% v/v)* 80% ACN HFBA (0.345% v/v)* Uniformity of response (RSD) 12% 14% 15% 18% 13% 12% 14% 13% 23% 30% * Acids FA, TFA and HFBA molar concentration each 26.5mM Page 112 of 246

114 Fig. 4.6a. Effect of organic solvent content on (a) CAD peak area (FIA, mobile phase 10 95% ACN, other conditions as per Fig. 4.2). Page 113 of 246

115 Fig. 4.6b. Effect of organic solvent content on (b) signal to noise ratio (FIA, mobile phase 10 95% ACN, other conditions as per Fig. 4.2). Page 114 of 246

116 Fig. 4.6c. Figure 4.6. Effect of organic solvent content on (c) noise (FIA, mobile phase 10 95% ACN, other conditions as per Fig. 4.2). Figure 4.6. Effect of organic solvent content on (a) peak area, (b) signal to noise ratio and (c) noise (FIA, mobile phase 10 95% ACN, other conditions as per Fig. 4.2). Page 115 of 246

117 3.2.6 Effect of elevated temperature (FIA) The simpler Corona Ultra design allows thermostatting of the nebuliser from 18 to 35 C (with the objective only to prevent freezing when using normal phase solvents); the temperature of the evaporator tube is at ambient. Using HILIC mobile phase, the Veo had a temperature range of C for the evaporator tube. Fig. 4.7 shows CAD peak areas for the 29 test compounds at 30, 60 and 80 C at high acetonitrile content (90% ACN). The effect of elevated temperature on the noise was small (not shown). There is clearly no advantage in using high evaporation temperatures for the majority of solutes: signal drops off dramatically in many cases, due to volatilisation of the solute. Nevertheless, evaporation temperature can be a tool for distinguishing analyte from back-ground based on volatility, and it is possible that optimising this temperature in smaller increments (e.g. 5 C) could be beneficial in some cases. Compounds such as caffeine which give moderate CAD response show a dramatic reduction in response at higher temperatures. Xanthine derivatives theobromine and theophylline, which are structurally similar to caffeine, perform well at low temperatures, but also had low CAD response at temperatures of 60 C and above. The base procainamide maintains good CAD peak area at the elevated temperature of 60 C, whereas the base diphenhydramine shows a drop-off comparable to the xanthine derivatives. Procainamide is more hydrophilic than diphenhydramine, and this result suggested a possible relationship with solute log D values. Thus the data were plotted in order of increasing (more positive) log D for hydrophobic solutes from left to right. Solutes on the left side of the plot (negative log D values) maintained good CAD peak areas even up to 80 C. Solutes on the right side of the plot, (positive log D values), showed drastic reduction in peak area at higher temperatures. It is possible that the hydrophobic solutes Page 116 of 246

118 are lost more readily as the ACN evaporates first from the aqueous organic mixture, whereas hydrophilic solutes can be solvated by the remaining aqueous liquid. Fundamental aerosol studies by Reid et al. showed that hydrophilic/hydrophobic mixtures in aerosols can form a biphasic droplet (Reid et al. 2011). To gain thermodynamic stability, hydrophobic components form surface lenses (partially engulfed structures), due to the relative surface tensions of the two phases (Reid et al. 2011). Such a system can perhaps favour migration of hydrophobic components to the surface of aerosol droplets, and at high temperatures lead to their evaporation. The above confirms that low evaporation temperatures are required for optimal CAD performance, as loss of signal can be dramatic for a variety of solutes at higher temperatures. This effect was also observed with the equivalent salt-buffered mobile phase at 10% ACN (data not shown). Page 117 of 246

119 Figure 4.7. Effect of elevated temperatures on Veo response in order of log D ( ve on left, +ve on right) (FIA, mobile phase 90% ACN, other conditions as per Fig. 2). Log D values were the average from three software packages (see Section 2) Blue = 30 C,Red = 60 C, Green = 80 C. Page 118 of 246